Tanmay Singh
Aspiring AI Engineer
Final-year B.E. student in Artificial Intelligence and Data Science who builds AI software that solves practical problems. Skilled in Python, machine learning, deep learning, computer vision, and web development. Seeking an entry-level AI Engineer role.
About
I am an aspiring AI Engineer who builds AI software that solves practical problems, working mainly in Python along with TypeScript, JavaScript, and SQL.
I am skilled in machine learning, deep learning, computer vision, and web development. I also build with generative and agentic AI, using RAG, FAISS, Sentence-Transformers, Ollama, and AI agents.
My projects include a local RAG assistant that answers questions on research papers with page-level citations, reaching 97.1% answer accuracy on a 68-question test set; an AI buyer agent that completes Razorpay test payments with human approval above Rs. 3,000; a reinforcement learning agent that finds bugs in a Mario-style game on its own; and a hand-gesture game controller with 99.0% test accuracy.
Education
Bachelor of Engineering in Artificial Intelligence & Data Science
Technical Skills
Languages
AI & ML
Generative AI & Agentic AI
Frameworks & Libraries
APIs & Databases
Developer Tools
Projects
CNN-Based Gesture Controlled Pac-Man
- Python
- PyTorch
- OpenCV
- CUDA
- Pygame
- Trained a MobileNetV2 CNN on a 2,000-image HaGRID dataset to classify hand gestures in real time.
- Reached 99.0% test accuracy on the left, right, up, and down commands.
- Built an OpenCV webcam pipeline with CUDA-enabled PyTorch inference to turn gestures into game controls.
- Added confidence thresholding and temporal smoothing to keep the controls responsive.
- Made a Pac-Man-style Pygame game with ghost AI, power-ups, and levels that runs at about 60 FPS.
Glitch Hunter – AI Game Testing System
- Python
- Stable-Baselines3
- Gymnasium
- Flask
- Trained a PPO reinforcement learning agent for 16 million steps across 8 parallel environments.
- The agent plays a Mario-style Pygame game on its own for continuous gameplay testing.
- Detects movement, clipping, score, and coin bugs by checking game-state invariants while it plays.
- Saves a screenshot, GIF, and PDF report as evidence for every bug it finds.
- Built a live Flask and Socket.IO dashboard that streams gameplay, rewards, and detected glitches.
Razorpay Agentic Commerce
- TypeScript
- Next.js
- PostgreSQL
- Gemini API
- Razorpay API
- Built an AI buyer agent with Google Gemini that understands shopping requests in plain words.
- The agent proposes products that match the requirements the user sets.
- Server code, not the AI, validates prices, spending limits, inventory, and authorization before payment.
- Added a secure policy and human-approval step before every purchase.
- Integrated Razorpay Test Mode with payment tracking, retries, and audit logs, deployed on Vercel.
Academic Research Assistant
- Python
- RAG
- FAISS
- Ollama
- Flask
- Built a fully local RAG app that answers questions about research PDFs with page-level citations.
- Runs Llama 3.2 (3B) through Ollama, with no API keys or cloud services.
- Combines FAISS vector search and BM25 keyword search with cross-encoder re-ranking.
- On a real paper, raised correct answers from 65% to over 90%, and the top-ranked right passage from 44.4% to 83.3%.
- Flask web app with streamed answers, chat history, and offline OCR, answering in about 1 to 1.5 seconds.