DEV Community: U G Murthy

The Soul of an Agent: Why You Should Stop Worrying and Start Building

U G Murthy — Sun, 15 Feb 2026 04:10:14 +0000

Why You Should Stop Worrying and Start Building?

Hey there, young hustlers and code curious! 🚀 Tired of all the doom-scrolling on social media? "AI agents are gonna steal your job!" "Everything's automating away!" Chill out. These thoughts belong to U G Murthy, paraphrased by AI – and here's the real tea: AI agents aren't the villain in your story. They're your sidekick, ready to level up your game. The future isn't about fearing them; it's about mastering how to work with them. That's the superpower skill that'll keep you ahead as AI models get smarter and cheaper every day. Think Clawdbot or Moltbot going viral, or Claude CoWork shaking up stocks – more game-changers are coming. Time to gear up and build!

What's an AI Agent, Anyway? (The Fun Spectrum)

Picture agents like levels in your favorite video game – from noob to boss mode:

Simple Agents: Just a quick chat or a chain of convos. Like texting a buddy for advice.
Moderate Agents: They search the web, pull out info, analyze it. Boom, instant research ninja!
Epic Agents: All that plus tools, workflows, and magic. They plan, execute, and adapt on their own.

Agents don't judge – they don't know you're "stupid" (your words, not mine!). They're always down to help, explain stuff, brainstorm wild ideas, teach you new tricks, or consult like a pro. Bonus: They mutate and evolve solo, and you can make 'em play any role – chef, coder, storyteller, you name it.

Your Secret Sauce: Help Your Agent Win (And Watch Magic Happen)

Agents are smart, but they're not mind-readers. Your success = their success. Here's how to team up like pros:

Build the vibe: Show 'em relevant context. "Hey, check this data..."
Steer the ship: "Consider this angle or that one." Give direction!
Spill your goals: "Do this exact thing for me."
Spark options: "What if we try A, B, or C?"
Get reflective: Ask questions like, "Why this over that?" to make 'em think deeper.

Nail this, and you're unstoppable.

Why Bother? The Perks Will Blow Your Mind

Working with agents? It's a cheat code for life:

Fresh ideas galore: Exposure to solutions you never dreamed of.
Ultimate sounding board: Bounce thoughts, reflect, and grow sharper.
Learn at warp speed: Pick up skills faster than binge-watching tutorials.
Playtime vibes: The journey's the real win – experiment, fail fast, laugh it off.
Unlock new worlds: Dive into stuff you never knew existed.
Bravery boost: With an agent in your corner, take risks, build the unknown. No fear!

My Wild Ride (U G Murthy's Story – Straight Fire)

I retired in 2016 with dusty programming skills from my youth – off coding for 20+ years. But hey, time on my hands? I dove back in, learned fresh, and loved it. Fast-forward to 2025: Coding agents popped up. I tried 'em all – CoPilot, Bolt, Cursor, Windsurf, Kilo Code (solid for a bit), then boom, AmpCode stole my heart.

Got hyped from blogs by Lex Fridman and Dwarfish, but my fave? The Latent Space podcast The AI Engineer Podcast – that's where agents hooked me. By 2026, I was building one without a plan. Every step sparked new ideas, like prizes for grinding the last level. Sleepless nights? Sure. But waking up buzzing with vibes? Priceless. Slowly, I shipped desiAgent, an SDK for agents – mostly powered by AmpCode, which I used to paraphrase this post.

This Lex Fridman Podcast hit me mid-walk, and was the inspiration for doing this post.

My thoughts 💭 - the prompt used for this post

The following is a list of thoughts for an essay - I want you convert this to a an essay jovial to natural tone easily readable and targeted mostly to youngster. Write the final essay to soul-of-an-agent-v1.md

The inspiration for writing this has come from  a podcast: https://podcasts.apple.com/in/podcast/lex-fridman-podcast/id1434243584?i=1000749366733  and lesson from my personal jouney 

Ensure you mention somewhere : that the though are mine (U G Murthy) Paraphrased by AI

### Soul of an Agent

### Key Message to drive home

AI Agents are not all gloom and doom as being discussed on most social media. Jobs will vanish etc. Gear your self develop a very important skill -  How work with agent and this requires us to understand what an agent is and how should I go about build my skill to use Agent. The AI models landscape is changing fast both in terms of capability and reducing cost. But its core is till the same allowing individuals to creating interesting use cases the one to go most viral is Clawdbot or Moltbot now and the other that crashed tech stocks is Claude CoWork and there will be more.

### Outline:

Following is an outline that needs to paraphrase and arrange in a logical flow

- Spectrum of agents
        - Simple - 1 chat conversation, chain of conversations 
        - Moderate - Search, extract, Analyse 
        - All of above + more tools + Agent workflows
    - Agent does not know I am stupid
    - Always there to help, explain, brainstorm, teach, consult
    - Agent can be mutate on its own
    - Agent can play different roles
- Every Agent needs help to succeed - your success lies in there
    - Show the the agent - build relevant context
    - Consider this and that - give it some direction
    - Do this… - expose your goal
    - Provide options - make it think about alternatives
    - Ask questions - Help the agent reflect
- How does it benefit me?
    - Exposure to new solutions
    - You get a sounding board making you reflect
    - Learn on the way and that too very quickly
    - Play - journey is more imp that end goal
    - Learn new things
     -  Don’t forget you have an agent to help you - be brave , take risks in building the unknown

- Personal journey 
    - I have been building ever since I retired in 2016. Though I had programming experience when I was younger, I was off coding for more than 2 decades. I had all the time in the world to learn - so I did
    - Glad I pursued learning to code. 2025, coding agents started to appear on the horizon and I embraced it -built and  Ditched many projects. Tried  Co Pilot, bolt, cursor, windsurf, settled on kilo code for a couple months and then I discovered ampCode   
    - Got inspired by many blog posts from Lex Freidman, Dwarfish,
    - favrouite podcast remains https://www.latent.space/podcast This is where I got curious about agents
    - 2026 I just started building an agent without a plan or a direction - every step in this direction gave birth to new ideas, it almost felt like a prize for the effort of previous step - there were sleepless nights but the journey was fun. I woke up every day with new ideas buzzing, it was hard to implement but slowly and steadily I build it
    - Here is desiAgent an SDK for agents : Most of the work was Ampcode, I review and understand its core.
        - Sharing my high level learnings of working fearlessly with agents by understanding their Soul
    - I got to writing this outline for this article while on a walk listening to Lex Friedman’s pod OpenClaw - https://podcasts.apple.com/in/podcast/lex-fridman-podcast/id1434243584?i=1000749366733


- Health warning:
    - AI Agents can consume you - beware of how much time and money you spend here
    - Do not take your eye of from - if you want to enjoy the future
        - The number of hours you sleep
        - Taking action to stay fit
        - 

- Conclusion
    - If you are young or retired dev wakeup / shapeup and start building stuff -

Health Warning: Don't Let It Own You ⚠️

Agents are addictive AF – time and cash can vanish. Keep your eyes on the prize for that epic future:

Sleep those 8 hours, no excuses.
Hit the gym, stay fit – action over screens.

Balance or burn out, fam.

Wrap-Up: Wake Up, Shape Up, Build Now!

Young guns or retired devs – this is your call to action. Stop scrolling, start shipping. Grab an agent, team up, and build cool stuff. The world's your playground. What's your first project? Go wild! 🌟

How to Build an Agent (in JavaScript)

U G Murthy — Sun, 19 Oct 2025 13:02:35 +0000

Disclaimer: I am not affiliated with AmpCode in any way. This content is created solely as a user of AmpCode's free mode.

Inspired by Thorsten Ball's article on building a Golang agent

This article, including all code generation, implementation, and debugging was by Amp Free

It’s not that hard to build a fully functioning, code-editing agent.

It seems like it would be. When you look at an agent editing files, running commands, wriggling itself out of errors, retrying different strategies - it seems like there has to be a secret behind it.

There isn’t. It’s an LLM, a loop, and enough tokens. It’s what we’ve been saying. The rest, the stuff that makes agents so addictive and impressive? Elbow grease.

But building a small and yet highly impressive agent doesn’t even require that. You can do it in less than 400 lines of code, most of which is boilerplate.

I’m going to show you how, right now. We’re going to write some code together and go from zero lines of code to "oh wow, this is… a game changer."

Here’s what we need:

Node.js
An OpenRouter API key that you set as an environment variable, OPENROUTER_API_KEY

Pencils out!

Let’s dive right in and get ourselves a new Node.js project set up:

mkdir js-coding-agent
cd js-coding-agent
pnpm init -y
pnpm add openai

Now, let’s create our files: index.js, agent.js, and tools.js.

First, agent.js:

const OpenAI = require("openai");

class Agent {
  constructor(client, getUserMessage, tools) {
    this.client = client;
    this.getUserMessage = getUserMessage;
    this.tools = tools;
  }

  async run() {
    const conversation = [];
    console.log("Chat with Claude (use 'ctrl-c' to quit)");

    let readUserInput = true;
    while (true) {
      if (readUserInput) {
        const userInput = await this.getUserMessage();
        if (!userInput) break;
        const userMessage = { role: "user", content: userInput };
        conversation.push(userMessage);
      }

      const message = await this.runInference(conversation);
      conversation.push(message);

      let toolResults = [];
      if (message.content && typeof message.content === "string") {
        console.log(`\x1b[93mClaude\x1b[0m: ${message.content}`);
      }

      if (message.tool_calls) {
        for (const toolCall of message.tool_calls) {
          const result = await this.executeTool(
            toolCall.id,
            toolCall.function.name,
            JSON.parse(toolCall.function.arguments)
          );
          toolResults.push(result);
        }
      }

      if (toolResults.length === 0) {
        readUserInput = true;
      } else {
        readUserInput = false;
        conversation.push(...toolResults);
      }
    }
  }

  async runInference(conversation) {
    const openaiTools = this.tools.map((tool) => ({
      type: "function",
      function: {
        name: tool.name,
        description: tool.description,
        parameters: tool.inputSchema,
      },
    }));

    try {
      const response = await this.client.chat.completions.create({
        model: "anthropic/claude-3-5-sonnet-20241022",
        messages: conversation,
        tools: openaiTools,
        tool_choice: "auto",
        max_tokens: 1024,
      });

      return response.choices[0].message;
    } catch (error) {
      throw new Error(`API Error: ${error.message}`);
    }
  }

  async executeTool(id, name, input) {
    const tool = this.tools.find((t) => t.name === name);
    if (!tool) {
      return {
        role: "tool",
        tool_call_id: id,
        content: "Tool not found",
      };
    }

    console.log(`\x1b[92mtool\x1b[0m: ${name}(${JSON.stringify(input)})`);
    try {
      const response = await tool.function(input);
      return {
        role: "tool",
        tool_call_id: id,
        content: response,
      };
    } catch (error) {
      return {
        role: "tool",
        tool_call_id: id,
        content: error.message,
      };
    }
  }
}

module.exports = Agent;

Yes, this compiles. But what we have here is an Agent that has access to an OpenAI client (configured for OpenRouter) and that can get a user message by reading from stdin on the terminal.

Now let’s add the missing index.js:

const OpenAI = require("openai");
const readline = require("readline");
const Agent = require("./agent");
const tools = require("./tools");

if (!process.env.OPENROUTER_API_KEY) {
  console.error("Please set OPENROUTER_API_KEY environment variable");
  process.exit(1);
}

const client = new OpenAI({
  baseURL: "https://openrouter.ai/api/v1",
  apiKey: process.env.OPENROUTER_API_KEY,
});

const rl = readline.createInterface({
  input: process.stdin,
  output: process.stdout,
});

function getUserMessage() {
  return new Promise((resolve) => {
    rl.question("\x1b[94mYou\x1b[0m: ", (input) => {
      const trimmed = input.trim();
      if (trimmed === "") resolve(null);
      resolve(trimmed);
    });
  });
}

async function main() {
  const agent = new Agent(client, getUserMessage, []);
  try {
    await agent.run();
  } catch (error) {
    console.error("Error:", error.message);
  } finally {
    rl.close();
    process.exit(0);
  }
}

main();

And tools.js as an empty array for now:

const tools = [];

module.exports = tools;

Let’s run it:

export OPENROUTER_API_KEY="your_api_key_here"
pnpm start

Then you can just talk to Claude, like this:

$ pnpm start

Chat with Claude (use 'ctrl-c' to quit)

You: Hey! I'm [Your Name]! How are you?

Claude: Hi [Your Name]! I'm doing well, thanks for asking. It's nice to meet you. How are you doing today? Is there something I can help you with?

You:

Notice how we kept the same conversation going over multiple turns. It remembers my name from the first message. The conversation grows longer with every turn and we send the whole conversation every time.

Okay, let’s move on, because this is not an agent yet. What’s an agent? An LLM with access to tools, giving it the ability to modify something outside the context window.

A First Tool

An LLM with access to tools? You send a prompt to the model that says it should reply in a certain way if it wants to use "a tool". Then you, as the receiver, "use the tool" by executing it and replying with the result.

To make (1) easier, the big model providers have built-in APIs to send tool definitions along.

Okay, now let’s build our first tool: read_file

The `read_file` tool

In order to define the read_file tool, we’re going to use JSON schemas, but keep in mind: this will all end up as strings that are sent to the model.

Each tool requires:

A name
A description
An input schema (JSON schema)
A function that executes the tool

So let’s add that to our tools.js:

const fs = require("fs/promises");

const readFileSchema = {
  type: "object",
  properties: {
    path: {
      type: "string",
      description: "The path to the file to read",
    },
  },
  required: ["path"],
};

async function readFile(input) {
  const { path } = input;
  try {
    const content = await fs.readFile(path, "utf-8");
    return content;
  } catch (error) {
    throw new Error(`Failed to read file: ${error.message}`);
  }
}

const tools = [
  {
    name: "read_file",
    description: "Read the contents of a file at the given path.",
    inputSchema: readFileSchema,
    required: ["path"],
    function: readFile,
  },
];

module.exports = tools;

Nothing fancy here. It’s a single async function, readFile, and descriptions the model will see.

Now, update index.js to use the tools:

const tools = require("./tools");
const agent = new Agent(client, getUserMessage, tools);

Time to try it!

$ pnpm start

Chat with Claude (use 'ctrl-c' to quit)

You: what's in index.js?

Claude: I'll help you check what's in the index.js file. Let me read it for you.

tool: read_file({"path":"index.js"})

Claude: Based on my review, index.js implements the main entry point for a Node.js CLI application that creates an AI agent. Here's a summary: - Imports necessary modules including OpenAI client, readline for CLI input, and custom Agent and tools - Checks for the OPENROUTER_API_KEY environment variable - Configures the OpenAI client to use OpenRouter's API endpoint - Sets up a readline interface for user input with colored prompts - Defines an async function to get user input - Initializes the Agent with the client, input function, and tools - Runs the agent in a try-catch block for error handling - Closes the readline interface and exits gracefully The code essentially bootstraps a conversational AI assistant that can interact with the user via command line and potentially use tools for additional functionality.

You:

It can read files now!

The `list_files` tool

If you’re anything like me, the first thing you do when you log into a new computer is to get your bearings by running ls.

Let’s give Claude the same ability:

const listFilesSchema = {
  type: "object",
  properties: {
    path: {
      type: "string",
      description:
        "Optional path to list files from, defaults to current directory",
    },
  },
  required: [],
};

async function listFiles(input) {
  const dir = input.path || ".";
  try {
    const entries = await fs.readdir(dir, { withFileTypes: true });
    const files = entries.map((entry) => {
      return entry.isDirectory() ? `${entry.name}/` : entry.name;
    });
    return JSON.stringify(files);
  } catch (error) {
    throw new Error(`Failed to list files: ${error.message}`);
  }
}

Add to tools array:

{
  name: 'list_files',
  description: 'List files and directories at the given path. If no path provided, lists current directory.',
  inputSchema: listFilesSchema,
  required: [],
  function: listFiles,
},

Let’s ask Claude what it can see in this directory.

$ pnpm start

Chat with Claude (use 'ctrl-c' to quit)

You: what do you see in this directory?

Claude: I'll help you see what's in the current directory. Let me list the files and directories for you.

tool: list_files({})

Claude: I can see several files and directories in the current directory: - README.md - agent.js - how-to-build-an-agent-js.md - index.js - node_modules/ - package.json - plan.md - tools.js - .gitignore

You:

Works! It can list the directory.

Let it `edit_file`

The last tool is edit_file — a tool that lets Claude edit files.

const editFileSchema = {
  type: 'object',
  properties: {
    path: {
      type: 'string',
      description: 'The path to the file to edit',
    },
    old_str: {
      type: 'string',
      description: 'The string to search for and replace. If empty, creates a new file.',
    },
    new_str: {
      type: 'string',
      description: 'The replacement string',
    },
  },
  required: ['path', 'old_str', 'new_str'],
};

async function editFile(input) {
  const { path, old_str, new_str } = input;
  try {
    let content = await fs.readFile(path, 'utf-8');
    if (content.includes(old_str)) {
      content = content.replace(old_str, new_str);
      await fs.writeFile(path, content, 'utf-8');
      return 'File edited successfully';
    } else if (old_str === '') {
      await fs.writeFile(path, new_str, 'utf-8');
      return 'File created successfully';
    } else {
      throw new Error('old_str not found in file');
    }
  } catch (error) {
    throw new Error(`Failed to edit file: ${error.message}`);
  }
}

{
  name: 'edit_file',
  description: 'Edit a file by replacing old_str with new_str. If old_str is empty, creates a new file.',
  inputSchema: editFileSchema,
  required: ['path', 'old_str', 'new_str'],
  function: editFile,
},

Let’s tell Claude to create a new FizzBuzz function.

$ pnpm start

Chat with Claude (use 'ctrl-c' to quit)

You: hey claude, create fizzbuzz.js that I can run with Nodejs and that has fizzbuzz in it and executes it

Claude: I'll create a fizzbuzz.js file that you can run with Node.js. First, let me check if the file already exists:

tool: list_files({})

Claude: I don't see a fizzbuzz.js file, so I'll create one with a classic FizzBuzz implementation. This program will print numbers from 1 to 100, but for multiples of 3 it will print "Fizz", for multiples of 5 it will print "Buzz", and for numbers that are multiples of both 3 and 5 it will print "FizzBuzz".

tool: edit*file({"path":"fizzbuzz.js","old_str":"","new_str":"/\**\\n _ FizzBuzz implementation in JavaScript\\n _ Prints numbers from 1 to 100, but:\\n _ - For multiples of 3, print \\"Fizz\\" instead of the number\\n _ - For multiples of 5, print \\"Buzz\\" instead of the number\\n \_ - For multiples of both 3 and 5, print \\"FizzBuzz\\"\\n \*/\\n\\nfunction fizzBuzz(n) {\\n for (let i = 1; i <= n; i++) {\\n let output = '';\\n \\n if (i % 3 === 0) output += 'Fizz';\\n if (i % 5 === 0) output += 'Buzz';\\n \\n console.log(output || i);\\n }\\n}\\n\\n// Execute FizzBuzz for numbers 1 to 100\\nconsole.log('Running FizzBuzz:');\\nfizzBuzz(100);"})

Claude: I've created a fizzbuzz.js file that you can run with Node.js. The program: 1. Defines a function called `fizzBuzz` that takes a parameter `n` for the upper limit 2. Loops from 1 to n (set to 100 in the execution) 3. For each number: - If it's divisible by 3, it adds "Fizz" to the output - If it's divisible by 5, it adds "Buzz" to the output - If it's divisible by both, it adds both resulting in "FizzBuzz" - If it's not divisible by either 3 or 5, it outputs the number itself 4. The script automatically executes the function for numbers 1 to 100 You can run this script using Node.js with the following command in your terminal: node fizzbuzz.js This will output the FizzBuzz sequence from 1 to 100 to your console.

Right?! It’s impressive, isn’t it? And that’s the most basic implementation of an agent you can probably come up with.

Isn’t this amazing?

If you’re anything like the engineers I’ve talked to, chances are that, while reading this, you have been waiting for the rabbit to be pulled out of the hat. But it’s not.

This is essentially all there is to the inner loop of a code-editing agent. Sure, integrating it into your editor, tweaking the system prompt, giving it the right feedback at the right time, a nice UI around it, support for multiple agents, and so on — we’ve built all of that in Amp, but it didn’t require moments of genius. All that was required was practical engineering and elbow grease.

These models are incredibly powerful now. 300 lines of code and three tools and now you’re to be able to talk to an alien intelligence that edits your code.

That’s why everything’s changing.

This article, including all code generation, implementation, and debugging, was by Amp Free.

This implementation was built in this AmpCode thread.

Inspired by Thorsten Ball's original article on building agents in Golang.

Code repo on github

Github Stats on your Github profile page

U G Murthy — Sat, 21 Dec 2024 00:42:39 +0000

Header picture credit: Grok2

Ever wondered how some GitHub profiles stand out with dynamic stats, vibrant visuals, and an interactive feel? Welcome to the art of creating a dynamic GitHub profile page! In this post, we'll dive into showcasing your coding prowess through GitHub statistics to turn your profile into a living resume of your coding journey. Let's make your GitHub profile not just a page, but an experience.

Create a GitHub Profile README:

GitHub allows you to create a special repository named after your username (e.g., yourusername/yourusername).

Navigate to GitHub's homepage and start creating a new repository. Set the name of the repository to your GitHub username.

Ensure the repository is initialised with a README.md file. This file will be displayed on your profile page.

Here is how you add Dynamic GitHub Stats:

GitHub Readme Stats:

Use services like github-readme-stats by Anurag Hazra to generate dynamic stats. You can include your GitHub contributions, stars, forks, and languages used. Here's how you can add it to your README:

Copy the Markdown link below, replacing yourname with your GitHub username:

[![yourname's GitHub stats](https://github-readme-stats.vercel.app/api?username=yourname)](https://github.com/yourname)

You can customise the appearance via query parameters - more details here.

Here is one example - adding icons and changing the theme

Copy the Markdown link below, replacing yourname with your GitHub username

[![yourname's GitHub stats](https://github-readme-stats.vercel.app/api?username=yourname&show_icons=true&theme=radical)](https://github.com/yourname)

Displaying Specific Stats:

Top Languages: Show the languages you most frequently use by adding:

[![Top Langs](https://github-readme-stats.vercel.app/api/top-langs/?username=yourname)](https://github.com/yourname)

Streak Stats: Include your coding streak with:

Copy the Markdown link below, replacing yourname with your GitHub username

[![GitHub Streak](https://streak-stats.demolab.com?user=yourname&border_radius=6)](https://github.com/yourname)

Profile Readme Generator Tools:

If you're looking for a more user-friendly way, tools like

can help you create a personalised README with various stats and styles. These tools provide templates and easy customisation options for your profile.

I am sure there is more you can showcase such as, list your top repos and the stats relating to them - check out GitHub Tracker in case you are interested.

That's it devs. By integrating these elements, you can enhance your GitHub profile to better showcase your coding activity and statistics. The exact customisation and choice of services depend on what you want to highlight on your profile.

SpeechTrack

U G Murthy — Sun, 24 Nov 2024 04:47:48 +0000

This is a submission for the AssemblyAI Challenge : 1) Sophisticated Speech to Text, 2) Really Rad Real-Time. and 3) No more Monkey Business

What I Built

SpeechTrack

Getting a deeper understanding of speech with AssemblyAI
Ever wondered how your speech sounds to others? Clarity and speed are the heartbeat of effective communication, but striking the perfect balance can be a challenge.

Enter SpeechTrack, your real-time speech companion.
With intuitive visual indicators for Clarity and Tempo, SpeechTrack empowers you to refine your delivery through post speech Feedback.

Whether you’re preparing for a big presentation, a podcast, or an important conversation,

SpeechTrack helps you speak with confidence, precision, and impact.

Try it now and transform the way you communicate!

Technical Overview

App - built using REMIX.run & React web framework tailwind.css and daisyUI for styling
App modules (at a high level)
1. Microphone - to stream Audio
2. Websocket module for receiving messages from AssemblyAI
3. AudioProcessor - to combine streamed Audio chunks to a .wav File object
4. Various React components to process & assemble received messages and maintain application state
5. Backend for front to deal with:
  - API keys and temporary tokens
  - Uploading audio,
  - Getting transcripts,
  - asking LeMUR to do evaluation of Speech
6. Finally present the Report in MARKDOWN format
7. Maintain a history of last 3 Speeches in localStorage on client side

Demo

Six minute video explaining how SpeechTrack works

Screenshot: Show's Server Logs while doing above video

Here are some use cases:

1. Learning from good speakers.

A good speaker can keep the audience spell bound. What is it that they do well that we can replicate. This inspiring speech is from the movie Coach Carter.

Here is the Speech Evaluation Report from LeMUR

Analysis & Feedback

Here's my analysis and feedback for the speech transcript in a human-readable Markdown format:

Speech Evaluation

Strengths

Effective Use of Pauses: With 8.75 pauses per minute (ppm), the speaker is within the ideal range of 5-10 ppm. This allows listeners to process key ideas and adds emphasis to important points.
Inspirational Content: The speech contains powerful, motivational messages about personal empowerment and positive influence on others.
Concise Delivery: At 104 words, the speech is relatively brief, which can help maintain audience attention and focus on core ideas.

Areas for Improvement

Speaking Speed: At 130 words per minute (wpm), the pace is slightly below the ideal range of 140-200 wpm. Increasing the speed slightly could make the delivery more engaging.
Filler Words: The use of "ah" was noted in the transcript. Reducing or eliminating filler words can enhance clarity and professionalism.
Structure: While the content is inspirational, a clearer structure with an introduction, main points, and conclusion could improve overall coherence.

Summary Feedback

The speaker delivers a powerful message with good use of pauses, allowing the audience to absorb the inspirational content. To enhance the speech:

Increase speaking speed slightly to reach the ideal range of 140-200 wpm.
Eliminate filler words like "ah" to maintain a smooth flow of ideas.
Consider structuring the speech more clearly, perhaps by grouping ideas into distinct sections.
The personal touch at the end ("Sir, I just want to say thank you. You saved my life.") is impactful but seems abrupt. Consider a smoother transition or integration of this personal element.

By implementing these suggestions, the speaker can elevate an already powerful message to create an even more impactful and polished presentation.

Screenshot: Show's Transcript

2. In-audio directives for Speech summary and sentiment Analysis.

The first line of audio is always checked for any directives (Summary, Sentiment Analysis) before speech->text transcript generation - see process details below.

In the following video hear the first line direct the app to prepare for summary and sentiment analysis.

Screenshot: Show's Realtime transcript with directives to parameter object

The various steps in the process Speech->Transript->Inference.

## All lines except those with leading ##'s are from server logs
## Detection of directive from first line
f(getComnand) from first FinalTranscript: I would like you to generate a catchy summary and sentiment analysis of the following story. {
  summarization: true,
  summary_model: 'catchy',
  summary_type: 'gist',
  sentiment_analysis: true
}

## The streamed audio packets are then combined with .wav header and
## file object for the wave file is uploaded 

url : https://api.assemblyai.com/v2/upload

## we then have a URL to the uploaded audio
f(fileUpload) : Uploaded Successfully  {
  upload_url: 'https://cdn.assemblyai.com/upload/ebabc86a-0a1d-4988-a6f0-5dc06352921d'
}
f(fileUpload): 472.023ms
/api/upload fileUpload: 540.219ms

## NOTE: parameters for audioURL -> transcript
## disfluencies - so assemblyai can transcribing filler words
## Parameters below are computed for every speech and depend
## on first line - see below
## audio_start_from - ensure we don't include first line while asking 
/api/upload getTranscriptFromURL: 6.628s

f(getTranscriptFromURL) {
  audio: 'https://cdn.assemblyai.com/upload/ebabc86a-0a1d-4988-a6f0-5dc06352921d',
  disfluencies: true,
  summarization: true,
  summary_model: 'catchy',
  summary_type: 'gist',
  sentiment_analysis: true,
  audio_start_from: 10010
}

## Transcript id along with a CUSTOM PROMPT is submitted to leMUR for 
## specfic speech evaluation response.
## CUSTOM PROMPT contains:
## 1. Description of the Task evaluation details,
## 2. Required outline of the evaluation report
## 3. Acceptable standards of a good speech
## 4. App generated metrics (duration, words per min, pauses, etc)
## 5. Report format as markdown

/api/feedback  1129a02c-4a28-49ea-8fbf-cb55f89a0214
f(askLeMUR): 9.234s

Screenshot: Show's Transcription and Sentiment Analysis

CUSTOM PROMPT used with LeMUR

You are **SpeechEvaluator**, an expert in analyzing and 
providing constructive feedback on speech delivery. 
Your role is to assess the quality of a speaker's 
performance based on key metrics and provide 
actionable insights to help them improve. 
You will be given structured data from a speech analysis, 
including metrics like *Words per Minute (*wpm*)*, 
*Pauses per Minute (*ppm*)*, 
duration, and word count, along with a transcript. 

**Your Task**:  

1. **Analyze Metrics**: 
Evaluate the speaker's delivery based on provided quantitative 
data:  
- *wpm*: Assess speaking speed (ideal range: 140–200 wpm).
- *ppm*: Evaluate pause usage (ideal range: 5–10 ppm).    
- *duration*: Consider how metrics align with speech length.    
- *wc*: Check for verbosity or brevity based on context.  

2. **Assess Transcript**:   
- Identify clarity issues, such as filler words, redundancy, or lack of focus.     
- Highlight the effectiveness of pauses and transitions between topics.
- Comment on structure and coherence, ensuring ideas flow logically.  

3. **Provide Feedback**:     
- Highlight **strengths** (e.g., engaging delivery, appropriate speed).     
- Suggest **areas for improvement**, including specific and actionable tips.  

4. **Use Accessible Language**: 
- Ensure your feedback is clear and easy for the speaker to understand, 
even if they are new to public speaking or storytelling or presentations.  

### **Example Output**:  

#### **Strengths**  
- *Good Speaking Speed*: wpm is within the recommended range, making the delivery engaging and energetic.  
- *Use of Pauses*: Pauses are moderately frequent, allowing listeners time to process key ideas.  

#### **Areas for Improvement**  

- *Filler Words*: Repeated use of 'um', 'uh', 'like', 'you know', and 'I mean' distracts 
from the main message. Replace these with deliberate pauses.  
- *Structure and Clarity*: The speech lacks a clear structure and transitions, 
making it harder to follow. Organise thoughts into sections for better flow.  

#### **Summary Feedback**  

The speaker demonstrates confidence and an engaging pace but should focus on 
reducing filler words like 'uh', 'um', and structuring ideas more effectively. 
With these adjustments, the speech will feel more polished and impactful.  
Follow this format consistently to ensure feedback is constructive and actionable. 
DO NOT APOLOGIZE FOR INFORMATION NOT AVAILABLE FOR ABOVE ANALYSIS"

Below is the metrics from this Speech in JSON format. Use it to evaluate the speech.

Code Repository
SpeechTrack
Application issues

You get Error : Not Authorized inspite of providing an API key. Solution: Click Next Speech to get a new token
The Application is stuck with the Spinner on endlessly : This is a case of http: 504 : Bad gateway Vercel.com times out serverless functions in 10secs. Most time leMUR task takes more time. Vercel serverless does not wait for long. Best is to run the application locally for big audio files.

Journey

The Assembly AI Hackathon immediately caught my attention. Exploring their Playground gave me a basic understanding of their APIs, and soon I was diving deeper with httpie, testing almost all REST APIs. The Speech-to-Text Streaming API stood out as the most exciting, and I knew it would be the foundation of my project.

Building SpeechTrack

With the help of ChatGPT, I brainstormed ideas and finalised SpeechTrack, a tool to provide real-time feedback to speakers on:

Clarity, measured through confidence scores.
Tempo, tracked in words per minute.

The name was a nod to Assembly AI’s focus on unraveling speech, evolving from ChatGPT’s original suggestion of VoiceTrack.

Steps in the Journey

Exploring APIs: I familiarised myself with Assembly AI’s tools using httpie.
Ideation: Brainstormed (w/LLMs) and refined the concept for SpeechTrack.
Implementation:
- Used the Streaming API for real-time speech insights.
- Faced challenges with audio streaming and creating WAV files. Resources like the Assembly AI Realtime Transcription Example and @3kb-dev’s WAV file guide were lifesaver.

Prompt Coverage.

In my opinion SpeechTrack qualifies for all 3 challenge prompts as it ended up using:

Streaming Speech-to-Text - to provide realtime cues (providing the audio and real-time transcript)
Speech-to-Text - Transcribe Audio (with in-audio parameter detection)
Speech Understanding (using transcript of (b) and LeMUR to provide analysis and feedback on Speech)

Feedback for Assembly AI

The API's are excellent and can be better if the following were available (I hope I did not miss reading it in the documentation)

A way to download the audio file that was uploaded to Assembly AI
A Streaming client for LeMUR. I noticed my app working locally but failing when hosted on vercel.com with time outs or http 504's
Lastly, For LeMUR tasks I would like have an API similar to what OpenRouter provides as that will expand access to models and allow good control over what we can instruct the model to do.

Overall, I see a lot of possibilities, keep up the good work.

Reflections & Conclusion

Though the journey had its hurdles, completing SpeechTrack was immensely satisfying. Thanks to Assembly AI for their fantastic APIs, great support (thanks to Lee Vaughn of Support Engineering at Assembly AI) and the Dev community for their support. Here’s to more learning and creating in the future!

Personas - an Ai Assistant

U G Murthy — Sun, 14 Apr 2024 16:37:05 +0000

This is a submission for the Cloudflare AI Challenge.

What I Built and What it does

Personas is a AI assistant - a Web app demonstrating the ease with which one can build app using Cloudflare Pages Cloudflare Worker combined with Cloudflare AI which provides plethora of AI models for a wide range of tasks.

Try Persona
A 2 minute video demo

PERSONAS Showcases 3 personas

Story Writer that is kidSafe
Coding Assistant that can help a developer
Sentiment Analysis & Reasoning

The above personas can take voice or text as input and will respond based on Handrails provided by the system prompt for the respective models.

Story Writer generates stories appropriate for Kids under 10 years with the option of translating to German. One can also analyse sentiment of the input prompt and seek the reasoning behind the evaluation. [tasks: Text Generation, Transation, Voice to text]

Coding Assistant is a useful for generating small code snippets in most programming languages, including syntax highlighting. You can also ask it to explain how a particular piece of code works. I used it do some of the grunt work for this project including getting ideas for react components.[tasks: Code Generation, and voice Voice to text]

Sentiment Analysis & Reasoning This was to demonstrate a good system prompt can be leverage an existing model and reduce reliance on specialist models for sentiment analysis. Here we see the the same text generation model used on Story Writer producing structured output as a JSON - which will be pretty useful for decision making in a chain of tasks.[tasks: Structured text Generation, Voice to text]

Demo

Try Persona
A 2 minute video demo

My Code

Code on Github

Journey

I came to know about the challenge on 4th April. Initially, I wasn't planning on doing much considering I was new to Cloudflare and learning about it seemed a daunting task.

This introductory video and the the Cloufare Workers AI got me started in no time.

Development process

Tested ideas for system prompt using Cloudflare AI playground
User examples in AI tutortials to build end points
tested the end points using httpie and sometime curl
Built the frontend with REMIX-Cloudflare

Multiple Models and/or Triple Task Types

Four Models were used to achieve many tasks Story Generation, Translation, Code Generation and Sentiment analysis:

@hf/mistral/mistral-7b-instruct-v0.2 two tasks:
Story Generation,
Sentiment Analysis
@cf/thebloke/discolm-german-7b-v1-awq for language translation
@hf/thebloke/codellama-7b-instruct-awq for Code Generation
@cf/openai/whisper for audio to text transcription.

Some unsung heros (commands/tools/libraries) without which this would have been impossible:

wrangler npx wrangler pages deployment tail <deployment id>
https from httpie
zod, zodix js modules to do validation and parsing
and last but not the least lodash my fav goto for anything that you think is impossible.

Future Plans:

Extend Story Writer to be hands free - voice commands for stories : hey Monk, tell me a story a picture story
Do Sentiment Analysis before generating stories.
Create Custom Persona
Extend the App to leverage Local open source LLMs from ollama.ai
Exploring Cloudflare LORA support for training to improve Personas further

Learning Lots of things learnt over last 7/8 days. From not knowing what is Cloudflare to deploying - I am pretty happy with the outcome. I was proud of two components I implement
1) Component to catch streaming data
2) the audio component. Both were challenging

Many Thanks to Cloudflare and dev.to for giving us this opportunity.

A blogsite template using @11ty

U G Murthy — Wed, 29 Jul 2020 09:10:07 +0000

I was reading an article on Serverless, this lead me to few terms/technologies/concepts that I wasn't familiar ; Jamstack and static generators. Exploring this terms led me to an introductory video on Creating a blog. This encouraged me to build a customisable blogsite template with following features:

Customisable blogsite ( modify _data/site.js )
Two column site - left column acting as navigator to blog posts
uses @11ty to build
uses bootstrap 4 for styles
is responsive
uses highlightjs for code blocks
blog posts are markdown files (see samples: /posts/*.md )
search blog content (left navigator shows resulting posts)

Where can I try it out?

Here is the live site on netlify.com (images load a bit slowly - dont know why?)

How do I use this template?

Fork a copy of this repo to your github account and clone to your local machine

Install and build locally

 
$ cd ~/blosite # change directory to you blogsite
$ npm install
$ npx @11ty/eleventy --serve
# Browse locally to check if you can access the blogsite
# check the search function too.

Signup with www.netlify.com using your github account
Build this repo using netlify.com (when you signup - netlify.com will guide you - chose defaults and you are done)

If the first 4 steps were successful i.e. you are able to see your site on netlify then proceed with customisation

Customise it by modifying contents of _data/site.js
Modify index.njk, about.njk as needed
Write you posts in markdown. All posts will reside ins posts
Ensure you update variables in frontmatter especially date to date of posting (as YYYY-MM-DD )else it defaults to file creation date which will keep changing to build date and thus not preserving order of the blogs in left navigator.
git commit the changes if any and push it to origin
netlify.com will automatically build the external blogsite.

Your feedback will be most appreciated - thanks for taking the time to read

References

Attribution
Image in this post by: Cookie_studio / Freepik

vim: Buffers, Windows, Tabs - Part 1

U G Murthy — Tue, 09 Jun 2020 10:27:54 +0000

Any developer involved in a large software developement project is actively editing multiple source files. There are multiple factors that makes such a developer efficient in the task of editing multiple files. Generally speaking such efficiency depends on:

navigating from one active file to another seamlessly
and active viewing area

vim provides way to do this using buffers, windows and tabs. In this article we will focus on buffers and windows

Before we get to the details lets understand the terms buffers and windows in vim.

A buffer is the in-memory text of the file that was opened using vim. It may be visible or hidden. It is identified by a buffer number and has flags to indicate whether it is current, has been modified or not etc.
A window is a view port on a buffer whereas
A tab is a collection of windows

When vim opens a file, a buffer is associated with it.

vi server.js

Not every buffer is associated with a file. We will see an example later.

Once in vim. Enter the following command in normal mode

:ls

which will show buffer information as below:

1 %a "server.js" line 1 Press ENTER or type command to continue

The integer 1 refers to the buffer number
The %a is a flag indicating that this is the active buffer
"server.js" the filename and
line 1 indicate the current line on which the cursor is positioned

Lets now open two more buffers. While in vim you can open another file by:
:edit config.json

Lets also create a buffer without any file associated with it:
:enew
and list all the buffers using :ls vim command.
1 "server.js" line 1 2 # "config.json" line 1 3 %a "[No Name]" line 1 Press ENTER or type command to continue
Note: buffer number 3 has "no name" and is active.

Switching between buffers is easy. Use :bn where n is buffer number :b2 will switch to buffer 2, makin it active.
Switching between buffers will work if the changes to a buffer is saved.

To view all buffers enter the following command in vim
:ball

You will now see three horizonal windows. You can switch from one to another by followed by any of the arrow keys.

Try :ls to check all buffers are now active as they are all visible.

You can close the the current window by followed by 'c' or by entering :q command. The window closes but the buffer is still there.

for more information on vim buffers enter the following command in vim

:help buffers

Next week look out for more on windows and tabs in Part-2 of this post.

DEV Community: U G Murthy

The Soul of an Agent: Why You Should Stop Worrying and Start Building

What's an AI Agent, Anyway? (The Fun Spectrum)

Your Secret Sauce: Help Your Agent Win (And Watch Magic Happen)

Why Bother? The Perks Will Blow Your Mind

My Wild Ride (U G Murthy's Story – Straight Fire)

Health Warning: Don't Let It Own You ⚠️

Wrap-Up: Wake Up, Shape Up, Build Now!

How to Build an Agent (in JavaScript)

Pencils out!

A First Tool

The read_file tool

The list_files tool

Let it edit_file

Isn’t this amazing?

Github Stats on your Github profile page

Create a GitHub Profile README:

Here is how you add Dynamic GitHub Stats:

GitHub Readme Stats:

Displaying Specific Stats:

Profile Readme Generator Tools:

SpeechTrack

What I Built

SpeechTrack

Demo

Here are some use cases:

1. Learning from good speakers.

Analysis & Feedback

Speech Evaluation

Strengths

Areas for Improvement

Summary Feedback

2. In-audio directives for Speech summary and sentiment Analysis.

Journey

Building SpeechTrack

Steps in the Journey

Feedback for Assembly AI

Reflections & Conclusion

Personas - an Ai Assistant

What I Built and What it does

Demo

My Code

Journey

A blogsite template using @11ty

vim: Buffers, Windows, Tabs - Part 1

The `read_file` tool

The `list_files` tool

Let it `edit_file`