I'm an AI engineer building systems that reason, plan, and learn. Background in electrical engineering and full-stack dev.

• 🌍 Exploring Interpretability in AI, Vision-Language-Action (VLA) models, and World Models
• 🛠️ Experience in Agentic AI, Deep Learning, Reinforcement Learning, and LLM applications
• 🔬 Building end-to-end systems: from autonomous research pipelines to AI-powered stock analysis
• ⚡ Fun fact: Built an autonomous research agent that discovers papers, runs experiments in Docker sandboxes, and writes weekly reports!

Autonomous Research Pipeline | Claude API | Docker

Autonomous research system that discovers papers from arXiv and Semantic Scholar, analyzes them with AI, runs experiments in Docker sandboxes or Modal cloud GPUs, and generates weekly narrative reports.

• 🧠 Built a 5-stage pipeline: Ingestion → Synthesis → Experiments → Analysis → Reporting
• 🔗 Maintains a knowledge graph (NetworkX) with hybrid retrieval (BM25 + vector RRF)
• 🔒 Experiment safety gates: AST validation, Bandit scan, auto-fix, human approval, Docker isolation
• 📊 Statistical analysis with confidence intervals, t-tests, and Cohen's d
• 🖥️ Streamlit web UI, Typer CLI, and APScheduler daemon for automated research cycles

Tech Stack: Python • Claude API • ChromaDB • Docker • Modal • Streamlit • NetworkX • SQLite

AI-Powered Stock Analysis | ReAct Reasoning | LLMs

Local AI-powered stock analysis app with a ReAct reasoning loop, technical and fundamental analysis, portfolio P&L tracking, sector heatmaps, and real-time alerts.

• 🤖 ReAct reasoning loop with multi-provider LLM fallover (NVIDIA NIM → Claude → GPT)
• 💾 Persistent ChromaDB vector memory with JSONL fallback
• 📊 Scored recommendations (STRONG BUY to AVOID), sector screening, and earnings calendar
• 📰 Financial news aggregation with sentiment scoring
• 🎨 Fintech-style dark GUI built with PyQt and matplotlib charting

Tech Stack: Python • PyQt • ChromaDB • yfinance • LLMs • matplotlib

Deep Q-Network (DQN) | PyTorch

Built an intelligent Tetris agent using Deep Reinforcement Learning with custom reward shaping and GA-optimized features.

• 🧠 Implemented GPU-accelerated DQN with experience replay and target networks
• 📈 Achieved 3× improvement in average lines cleared vs. baseline
• 🎯 Engineered board-state features (aggregate height, holes, bumpiness) with delta-based rewards
• 📊 Trained for 1.5M timesteps with automated evaluation pipeline

Tech Stack: Python • PyTorch • Reinforcement Learning

Retrieval-Augmented Generation | FastAPI | LLMs

End-to-end RAG system for querying research papers with semantic search and grounded responses.

• 🔍 Implemented PDF parsing, semantic chunking, and vector-based retrieval
• 🤖 Built LLM-powered Q&A with source citations to reduce hallucinations
• ⚡ Designed FastAPI web interface + CLI for low-latency question answering
• 💾 Integrated ChromaDB for efficient vector storage and similarity search

Tech Stack: Python • LangChain • ChromaDB • FastAPI • Vector Embeddings

Northeastern University | Master of Science in Artificial Intelligence 📍 San Jose, CA | Sep 2025 – May 2027

APJ Abdul Kalam Technological University | Bachelor of Technology in Electrical & Electronics Engineering 📍 Kerala, India | Oct 2020 – May 2024