DEV Community: Shubhankar Madgulkar

I Built an AI Sidekick to Tame My Research Workflow - Seeking Feedback!

Shubhankar Madgulkar — Thu, 10 Apr 2025 13:37:27 +0000

Hey Dev.to community,

Like probably a lot of folks here, I spend a ton of time digging through information online. By day, I'm an equity research analyst, which means constantly sifting through news, reports, press releases, you name it. It's easy to get lost in the tabs and struggle to connect the dots later.

I kept thinking there had to be a better way than endless copy-pasting into notes or just hoping I'd remember where I saw that one crucial stat. So, I decided to build something about it.

I created Research Assistant, a Chrome extension designed to be an AI-powered sidekick for web research. The idea is simple: hit the extension button on any webpage, and it uses AI (you plug in your own OpenAI or Anthropic key) to give you summaries and key takeaways.

But the real goal was knowledge management. You can save these insights directly into a personal knowledge base within the extension. You can create different 'workspaces' for projects (e.g., Q1 Earnings, Competitor Analysis) and tag stuff automatically by topic. Later, you can search across everything you've saved.

Key things I focused on:

Smart Analysis: Get the gist quickly without reading everything top-to-bottom.
Personal Knowledge Base: Actually retain what you find, organized your way.
Workspaces & Tagging: Keep different research streams separate and find things easily.
Privacy First: This was crucial for me. All data stays local on your machine. No tracking, no cloud sync of your research content. It uses your API key to talk to the AI provider, but your saved notes don't leave your computer.

Why I'm posting here:

Honestly, this started as a tool for myself. Some folks in my office tried it and found it genuinely useful, which was encouraging. I even showed it to family members in school/college, and they seemed to like it too (though, you know, maybe they're just being nice!).

Now, I'd love to get some real, honest feedback from a broader tech-savvy community like this one. Is this actually useful? What's missing? Does it break on certain sites?

If you do a lot of online research, whether for work, learning, or personal projects, I'd be grateful if you'd give it a try and let me know what you think.

You can grab it here: Link to Chrome Webstore

Check out some of my other projects here.

Thanks for reading! Looking forward to any thoughts or suggestions.

I Got Carried Away Coding a Simple Lander Game... Now I Can't Even Beat Level 1

Shubhankar Madgulkar — Tue, 01 Apr 2025 20:47:51 +0000

Hey dev community,

Had one of those slower days at the office recently, you know the type. Got the itch to code something fun, just to keep the fingers moving. Decided on a classic lander game – nice and simple, right? The plan was straightforward: land on the Moon (Level 1), then Mars (Level 2), then Venus (Level 3). Easy peasy.

So I started coding and putting stuff together. And honestly? I started getting really into it. Tweaking the physics, adding little obstacles, adjusting the landing sensitivity... I was aiming for that satisfying retro challenge.

Fast forward a bit, and I felt like the whole thing was ready to go. Time for the first proper playthrough! I fire it up, ready to nail that landing and move on to Mars.

Except... I can't.

I genuinely cannot land the damn ship.

Somewhere in the tweaking and "applying physics", I turned the difficulty dial up to 11 without realizing it. Now I am not a gamer by any means, and yes I wanted it to be challenging, sure, but dammit i can't even get past the first level.

The irony is killing me. I coded this thing, and I'm completely locked out of levels 2 and 3 because I can't beat my own creation. I haven't even seen what Mars and Venus are supposed to look like in action yet!

Is it just me who is so bad and playing this?

Check out the game here

Feel free to check out my other silly projects [here].(https://www.shubhankarmadgulkar.com)

Ditch the Google Searches: Streamlining My Terminal Workflow with Spren (and Rust!)

Shubhankar Madgulkar — Tue, 11 Mar 2025 14:51:58 +0000

Let's be honest, my terminal workflow was a hot mess. Anyone else spend more time Googling commands than actually writing code? It wasn't just the typos (though, let's face it, those were a daily occurrence). It was the constant context switching: trying to remember complex commands, failing miserably, and then bouncing between Google searches and LLMs. Copy-pasting commands and hoping they worked? Not exactly efficient.

And then I realized it wasn't just me. New programmers, especially those coming from non-Unix backgrounds, were struggling even more. Breaking their flow to find the right commands was a huge productivity killer. I knew there had to be a better way.

Enter Spren: My LLM-Powered Terminal Assistant

That's when Spren was born. The idea was simple: bring the power of LLMs directly into the terminal, eliminating the need for constant context switching. Imagine asking your terminal, in plain English, for the command you need, and having it appear instantly. That was the dream.

Key Features (and Why I Built Them)

Direct LLM Integration (Why? To stay in the zone!): Get terminal commands directly from your preferred LLM (Claude, OpenAI, and more to come!), without leaving your terminal.
Cross-Platform Simplicity (Why? Because we all use different setups!): Works seamlessly on Windows (PowerShell, CMD), Linux (Bash), and macOS.
Easy Setup (Why? Because no one likes complex installations!): Just download the executable, and you're ready to go. No dependencies needed.
Flow Preservation (Why? To boost productivity, of course!): Eliminate workflow interruptions and stay focused on your tasks.
Empowering New Programmers (Why? Because the terminal shouldn't be intimidating!): Makes the command-line interface more accessible and user-friendly.

Flexibility is Key: Your LLM, Your Rules

Spren's core design principle is flexibility. I wanted a tool that could adapt to any user's workflow, regardless of their preferred LLM. That's why Spren supports both Claude and OpenAI, with the ability to easily plug in other LLMs in the future. You're not locked into a specific ecosystem; you choose what works for you.

Built with Rust: Performance and Safety First

Building Spren with Rust has been an incredible learning experience. Rust's performance and safety features allowed me to create a lean and efficient tool that gets the job done without any unnecessary overhead. Plus, it was a great way to deepen my Rust skills!

Diving into the Technical Details

Let's take a peek under the hood:

1. Configuration with `config.toml`

Spren uses a config.toml file for easy customization. Here's an example:

[ai]
provider = "anthropic"
anthropic_api_key = "your-key-here"
openai_api_key = ""
model = "sonnet-3.5"
max_tokens = 4000

[security]
require_confirmation = true
dangerous_commands = ["rm", "sudo"]

[display]
show_execution_time = true
color_output = true

[debug]
show_raw_response = false
log_level = "info"

This file allows you to set your AI provider, API keys, security settings, and display preferences.

2. Shell Detection and Command Execution: Adapting to Your Environment

Spren intelligently detects your shell environment, ensuring commands are formatted correctly. Here's how it works:

Automatic Detection: Spren determines your shell (Bash, PowerShell, or CMD) by examining environment variables and OS configurations.

Shell-Specific Formatting: Once the shell is identified, Spren formats commands using the appropriate syntax. For instance, PowerShell commands are wrapped with specific encoding and output formatting to handle diverse data types correctly:

// src/executor.rs
pub async fn execute_command(command: &str) -> Result<CommandOutput> {
    // ...
    let formatted_command = match shell_type {
        ShellType::PowerShell => {
            format!(
                "$OutputEncoding = [Console]::OutputEncoding = [Text.Encoding]::UTF8; \
                 $FormatEnumerationLimit = -1; \
                 $result = {}; \
                 if ($result -is [System.Array]) {{ \
                     $result | Format-Table -AutoSize -Wrap | Out-String -Width 120 \
                 }} elseif ($null -ne $result) {{ \
                     $result | Format-Table -AutoSize -Wrap | Out-String -Width 120 \
                 }} else {{ \
                     \"No output\" \
                 }}",
                command
            )
        },
        _ => shell_type.format_command(command)
    };
    // ...
}

Execution: Finally, Spren executes the formatted command in your detected shell, providing seamless integration with your existing workflow.

This ensures that Spren works consistently across different platforms and shells, without requiring manual configuration.

3. Security Measures

Spren includes security features like command validation and confirmation prompts for dangerous commands (defined in config.toml).

How to Get Started

Download Spren: Grab the executable from Get Spren.
Run it! Spren is designed to be plug-and-play.
Start asking questions! Use natural language to get the commands you need.

Let's Build the Spren Community!

I'd love to hear your feedback and see how Spren helps you streamline your workflow. If you're tired of breaking your flow or if you're a new programmer looking for a helping hand, give Spren a try.