Full-Stack Developer | AI/ML Engineer | Building data-efficient, interpretable AI solutions

Welcome to my GitHub where I build AI systems that work with real-world constraints (limited data, need for transparency, immediate deployment).

Repository | Live Demo

A production-grade quality inspection system demonstrating interpretable ML without massive datasets. This project showcases cutting-edge thinking about agentic AI and why transparency matters in high-stakes domains.

What makes this special:

• ✅ Agentic Architecture — Dynamic tool selection & decision-making workflows
• ✅ LLM Integration — OpenAI, Google Gemini, Claude, with graceful fallbacks
• ✅ Data Efficiency — Works with minimal samples; no training data required
• ✅ Full Interpretability — Every decision is traceable to specific measurements
• ✅ Production Ready — Deployed on Streamlit with comprehensive documentation

Tech Stack: Python, PyTorch, Streamlit, LLMs (OpenAI/Claude/Gemini), scikit-image
Real-world Impact: Shows how to build AI for agricultural & manufacturing quality control when data is scarce

Repository | Live Demo

A production-grade RAG pipeline designed for deep analysis of academic research. This tool bridges the gap between raw PDF data and actionable insights using a sophisticated hybrid retrieval architecture.

What makes this special:

• ✅ Hybrid Retrieval — Combines FAISS semantic search with BM25 keyword matching for maximum recall.
• ✅ Cross-Encoder Reranking — Prioritizes the most relevant paper segments using ms-marco-MiniLM to ensure high-quality context.
• ✅ Dual AI Engine — Flexibility between local Flan-T5 for privacy and Gemini 1.5 Flash for high-reasoning synthesis.
• ✅ Metadata Anchoring — Enforces strict grounding in paper Abstracts and Titles to minimize hallucinations.
• ✅ Optimized Workflow — Features parse-and-purge memory management and adaptive context windowing.

Tech Stack: Python, Streamlit, Google Gemini 1.5, FAISS, BAAI Embeddings, PyMuPDF4LLM
Real-world Impact: A powerful, privacy-focused RAG pipeline that bridges the gap between raw research PDFs and actionable scientific insights using hybrid search and dual AI engines.

🤖 AI Projects — Collection of production AI applications:

• 🎧 OmniConvert — All-in-one media converter (audio, text, images, video)
• 📚 Course Generator — Personalized course outlines with GPT-4 | Demonstrates prompt engineering & API integration
• 🤖 FAQ Chatbot — Production chatbot with FastAPI & OpenAI GPT-3.5 | Built conversation history management
• 🎁 Gift Recommendation System — AI-powered personalization engine

🧠 Machine Learning Projects — Diverse ML applications:

• 🩺 Diabetes Prediction — Health data classification with scikit-learn
• 🎵 Lyrics Generator — Generative model using LSTM networks
• 🧬 MBTI Personality Prediction — NLP-based personality classification

👁️ Computer Vision — OCR application with Streamlit & Flask

Languages: Python, JavaScript
ML/AI: PyTorch, TensorFlow, scikit-learn, LLM APIs (OpenAI, Claude, Gemini)
Frameworks: FastAPI, Streamlit, Flask, Spring Boot
Tools: Git, Jupyter, Docker, Figma, Regex101

Resources for mastering key technologies:

• 📦 Spring Boot Learning Path — My roadmap for Java Springboot
• ⚡ FastAPI Hands-On — Quick tutorial for API beginners
• 🛢️ SQL Tutorial — Interactive SQL fundamentals

Design & UI: Canva | Figma | CodePen
Development: Regex101

• Open source contributions in ML/AI and interpretable AI
• Collaborations on data-efficient, production-grade AI systems
• Roles combining full-stack development with ML engineering

I'd love to discuss AI, engineering challenges, or collaboration opportunities:

• 💼 LinkedIn — Let's connect professionally
• 🌐 Portfolio — See my work
• 📧 Email: moumitabasu597@gmail.com

Building AI systems that are fast, interpretable, and production-ready. 🚀