DEV Community: Agung Sidharta So

MindsDB Stress Test Knowledge Base - My Journey Creating App using MindsDB KB

Agung Sidharta So — Sat, 28 Jun 2025 21:00:11 +0000

About MindsDB Stress Test

I found this quest on Quira with MindsDB Knowledge Base (KB) as the topic of the quest. I have been interested about MindsDB for some time but with quest and prize as cherry on top, why not.

Idea

I need to find idea before starting to write the app and then decided to create hackathon aggregator web app that includes a chatbot to showcase capability of semantic search using MindsDB KB. Why hackathon aggregator ? Simply because I love to participate in hackathon.

Development

I started using vibe coding, in order to help me accelerate development especially on creation of front-end. To interact with MindsDB, I don't rely on MindsDB SDK but solely rely on mysql SDK in Node.js, in this case using mysql2 library. For IDE, I use cursor.

Challenge

While doing development and learning at same time, I found issues that perhaps could be easily avoided if I use MindsDB SDK, as the issue seems like happening only if I use mysql client to access MindsDB using its MySQL API. I'll write more detail about the issue and submit as bugs on MindsDB repo.

Repository

Please find the repository of the app here.

Demo Video

It's included in the github repo. Please watch it there.

Result

Roughly 2 days doing vibe coding and learning by doing, I can create a hackathon aggregator web app with following feature:
🔍 Smart Search & Filtering: Find hackathons by title, description, tags, status, and type
🤖 AI-Powered Chatbot: Interactive assistant that helps users find relevant hackathons
🎯 Personalized Recommendations: AI-driven suggestions based on user questions
📱 Responsive Design: Beautiful, modern UI that works on all devices
🔄 Multi-Source Integration: Aggregates data from Devpost, Topcoder, and Quira
⚡ Real-time Updates: Hackathon data fetching and updates
🔐 Admin Dashboard: Manage data sources

Recommendation

Try to use MindsDB SDK to do development on MindsDB instead of mysql client. While I have not tested it, theoretically the development should be smoother with native SDK.
Accelerate your development with AI (vibe coding).
Understand how AI Model and Agent work inside MindsDB. There'll be no need to use another AI Agent framework, let model and agent work from your SQL statement. Without proper understanding, I originally would like to use another agent framework such as langgraph.

From Youtube to Blog Post - use agent.ai to write blog out of youtube video

Agung Sidharta So — Sun, 26 Jan 2025 18:38:12 +0000

This is a submission for the Agent.ai Challenge: Productivity-Pro Agent (See Details)

What I Built

Imagine you are a youtube content creator and you are so good at making video and attracting your audiences, however after long days you still need to write a blog post explaining the content you make to extend your outreach. Or you are following a blog writing competition based on certain topic and you find this youtube video that could tick your reader but you have difficulty where to start.

Introducing Youtube to Blog Post built using agent.ai.

Demo

Demo Video:

Link to Agent: here

~~I plan to add full automation using Medium API before submission time (if time permits).~~ Unfortunately, Medium no longer allows to create new integration token, so the blog post output needs to be manually published on Medium.

Agent.ai Experience

Very easy and smooth, building AI agent with builder is straightforward. I follow the course Intro to Agent Builder which explains step by step how to build AI agent and then try to do it myself. Originally I thought I will need to create a code to get youtube transcription, but it has been well provided by agent builder, so what is left for me is designing the prompt for LLM. ~~An out-of-the-box Create Output action to directly write to Medium (platform I chose for blog) is not there yet but it should be solved by writing serverless function or invoking web API.~~

Resume AI - Polish your Resume for New Beginning or Fresh Start - Powered by Taipy, Langchain, Gemini, and Github Copilot

Agung Sidharta So — Sun, 12 Jan 2025 12:31:47 +0000

This is a submission for the GitHub Copilot Challenge: New Beginnings or Fresh Start

What I Built

Everyone deserves a new beginning or fresh start in their career journey, especially as we just entered new year of 2025. In this opportunity, I built a modern resume analysis tool that combines the power of Google's Gemini AI with Taipy's intuitive UI framework. The application helps users give insights, check grammar, propose improvement on their resume, and also interact with the resume through a chat with AI interface. I sincerely hope this could help someone who would like to look for new role and/or transition career.

Technical Stack

Frontend: Taipy UI
Backend: Python (Taipy)
AI Model: Google's Gemini 1.5 Flash
LLM Orchestration: Langchain

Key Features

1. Resume Analysis Dashboard

The application provides instant metrics including:

Skills distribution (Technical vs Soft skills)
Years of experience
Role analysis (Individual vs Managerial)
Company history
Interactive WordCloud visualization for extracted skills

2. Grammar Check & Improvements

Automated grammar analysis
Style suggestions
Keyword optimization
Structure improvements

3. AI Chat Interface

Context-aware conversations
Resume-specific queries
Professional recommendations

Demo

Link to the app: here.

When app is shown, simply upload your resume and after analysis is complete, 3 buttons based on Key Features will show. Also, please visit github link for more detail about sample questions to chat with the AI.

All these are running on free tier. If you get hit by rate limit of Google AI API, please try again. If the web app doesn't show up immediately, the instance may be deactivated due to inactivity. Wait for a couple of minutes and refresh the app. Also, if it looks like it hangs up on you after message saying "Resume analysis complete!", just refresh the page, it should show up with results (this is perhaps due to low CPU on the free tier).

Repo

Gungz / taipy-resume-ai

Resume AI with Taipy and Gemini

An AI-powered resume analyzer that provides insights, grammar checks, and interactive chat capabilities using Google's Gemini model.

Features

📊 Resume Analysis Dashboard
🔍 Grammar Check & Suggestions
💬 Interactive Chat with Resume Context
📈 Visual Metrics & Word Cloud
📱 Responsive UI

Prerequisites

Python 3.10+
Google API Key
PyMuPDF (for PDF processing)
Taipy

Installation

Clone the repository:

git clone <repository-url>
cd <repository-name>

Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: .\venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

Set up environment variables: Create a .env file in the root directory:

GOOGLE_API_KEY=your_google_api_key_here

Get your Google API key from here

This project uses gemini-1.5-flash

Local Deployment

Start the development server:

taipy run main.py --use-reloader

Acccess the application at http://localhost:10000

Environment Variables

GOOGLE_API_KEY: Your Google API key for Gemini

Sample Questions for AI

What do you think of my…

View on GitHub

Copilot Experience

I used Copilot throughout development process:

Use the autocomplete of code proposed by copilot. This is very helpful to avoid writing boilerplate code. For example: When I created a variable for chart data of skills, copilot immediately recommended for autocompletion 2 other variables for chart data of roles and companies, simply because it has the context of what I would like to write.
When there's a mistake while writing code e.g. python indentation, missing comma when writing taipy code (as I am just learning), I can hover on the mistake and ask copilot to fix it for me.
What is most useful for me and I use it like 80% of the time is the copilot chat (using Claude Sonnet 3.5 - it works much better than GPT 4o). I use it to iteratively expand functionality in my code and also to write README doc. Few of my chats with copilot chat that I can remember:
- load pdf content langchain
- use langchain with prompt and use system prompt with ChatGoogleGenerativeAI
- ask llm to return age, gender, number of skills, year of experience, number of roles, number of company in json
- show the json result using appealing chart in taipy
- I want the metric shown as individual number but appealing
- Create page in taipy for resume insight, grammar check and potential improvement, and chat with resume
- I want analysis results part visible only after resume is uploaded
- generate wordcloud from skills

I still need to do some fixes from generated code but it helps me accelerate the development. IMHO, majority of the fixes that I need to do is because taipy as framework or library was still new and not many examples can be found easily when googling it, hence copilot chat was unable to provide right example.

Some notable mistakes that took time to fix:

Copilot proposed visible in attribute of taipy's part to control visibility while the real attribute for that is render.
Copilot proposed to use base64encoded value in taipy's image to render an image while what is needed is the byte value from buffer.
Copilot also didn't propose to use rebuild attribute in taipy's chart as it is needed in order to render chart after data change.

Despite all the mistakes, I strongly recommend using Copilot to accelerate code development especially if the usage is for framework or library more widely known.

GitHub Models

I don't use Github Models for this purpose.

Conclusion

I am happy I am able to learn new things in this 24 hours window (mostly taipy and langchain). I learn to be able to use Github copilot effectively. I also sincerely hope the app can be useful for someone looking for new beginning or fresh start in their career by improving their resume. This is also a good upskill for me in 2025 as Taipy looks promising for quick prototyping (and move to production) and it's really fast.

Glam Up My Markup: Camp Activities

Agung Sidharta So — Sun, 31 Mar 2024 14:45:49 +0000

This is a submission for DEV Challenge v24.03.20, Glam Up My Markup: Camp Activities

What I Built

I built aesthetic and interactive form for camp activities which will automatically fill the information counsellor need to know based on selection of camp activity (can still be modified in case user needs to add more information).

Demo

Journey

I learned that with the prompt and initial code given to us, AI can already generate the aesthetic, interactive, and usable form, which is really great starting point of development.

Chitato Rasa Mie Goreng CSS Art

Agung Sidharta So — Sun, 31 Mar 2024 14:30:14 +0000

This is a submission for DEV Challenge v24.03.20, CSS Art: Chitato Rasa Mie Goreng

Inspiration

Chitato Rasa Mie Goreng (see Cover Image)

Demo

Journey

I learned how to use box-shadow, linear-gradient background, and transform translate of CSS. I believe making the real Chitato pack from CSS is still too much for me but at least I try. AI also helps.

Fetch API - Your Modern Browser's HTTP Requests

Agung Sidharta So — Sun, 24 Mar 2024 16:47:18 +0000

This is a submission for DEV Challenge v24.03.20, One Byte Explainer: Browser API or Feature.

Explainer

The Fetch API is a modern way to make HTTP requests in JavaScript. It provides a clean, promise-based interface for fetching resources asynchronously. It simplifies tasks like fetching data from APIs, uploading files, and handling cross-origin requests.