Full Stack AI Platform
Tutoring AI
2 months
A full-stack AI platform that turns text prompts into educational animation videos by generating Manim scripts and orchestrating AI-assisted rendering workflows.
- Integrated Mistral AI, OpenAI, and Google Gemini to generate and refine Python Manim animation scripts.
- Built asynchronous video processing with Celery and Redis to handle concurrent rendering jobs efficiently.
- Containerized the application stack with Docker Compose for frontend, API, and worker services.
Python Library
Pocket Agent
3 weeks
A reusable Python library for agentic LLM task execution with a unified provider interface and MCP-driven tool integration.
- Built a plug-and-play interface supporting multiple LLM providers with backend integration.
- Implemented MCP client support for dynamic tool registration and asynchronous external execution.
- Added a structured plan-act-observe-respond workflow for reliable multi-step tool use.
Real-Time Multiplayer Game
Chess Online
4 weeks
A real-time multiplayer chess application with websocket matchmaking, canvas-based board rendering, and observability for distributed tracing and logs.
- Implemented websocket-based matchmaking to pair players and synchronize game state in real time.
- Built the interactive chess board using React Konva for canvas-based 2D rendering.
- Added an OpenTelemetry observability stack for tracing, metrics, and centralized logging.
Full Stack Web App
Campus E-Wallet
10 weeks
A campus-focused e-wallet system built for secure digital payments within the institute with auditable transaction flows.
- Designed a normalized SQL database for users, wallets, transactions, and audit logs.
- Implemented role-based authentication with ACID-compliant transaction management.
- Optimized concurrency, indexing, and query performance for secure wallet operations.
6 weeks
A SAR flood detection pipeline that analyzes Sentinel-1 data for remote-sensing workflows and model evaluation.
- Researched transformer-based geospatial foundation models for remote sensing tasks.
- Built preprocessing and evaluation steps for SAR flood segmentation workflows.
- Worked through satellite image preprocessing, model evaluation, and result interpretation.