I build end-to-end AI systems — from RAG pipelines and reinforcement-learning agents to computer vision and production-grade APIs.
🎓 MSc Data Science · PSG College of Technology | 💼 Open to full-time roles in Data Science, ML & AI Engineering
- 🤖 Applied ML & Deep Learning — vision, speech, anomaly detection, time-series forecasting
- 📚 LLM & RAG systems — retrieval pipelines, semantic chunking, faithfulness evaluation
- 🎯 Reinforcement Learning — Soft Actor-Critic agents in custom Gym environments, tracked with MLflow
- 🛠️ Full-stack delivery — FastAPI / Flask / Django backends, React frontends, Docker, PostgreSQL / PostGIS
- 🏆 Hackathon builder — shipped working products under deadline for GOAT Robotics, Katomaran, IncuBrix and more
|
One API to publish content to Discourse, Lemmy, Dev.to, Ghost and signed webhooks. Connector contract, rules engine, job runner, certification suite, Docker Compose stack and free-tier cloud deploy.
|
Autonomous BTC/USD trading agent using Soft Actor-Critic with a CNN encoder over candlestick images. Custom Gymnasium env, live Streamlit dashboard and full MLflow tracking.
|
|
Multi-document RAG chatbot with intent routing (Q&A vs summary), OCR fallback, two-column layout handling and NLI-based faithfulness scoring.
|
Flood-response command platform for Chennai: GIS risk zones, role-based access, live resource tracking for shelters, vehicles, personnel and supplies.
|
|
AI-generated image detector with Grad-CAM explainability, FGSM / blur / frequency attack testing and adversarial training for robustness.
|
AI public-speaking coach: three timed rounds scored on relevance, coherence, fluency and vocal emotion from Whisper transcripts and audio features.
|
|
Extracts fields and line items from scanned invoice PDFs, detects seals/signatures, and outputs a confidence-scored verifiability report.
|
Register faces, recognise them live from a webcam, and ask natural-language questions about registrations via a LangChain + FAISS pipeline.
|
|
Phone-first emergency PWA: automatic crash detection from GPS speed + accelerometer, voice SOS, Twilio SMS to family, and the nearest of 15,000+ hospitals, police and rescue services across Tamil Nadu.
|
Search and recommendation engine that ranks content by quality: Boolean/vector/BM25 retrieval, query expansion (nDCG 0.61 → 0.99), quality classification, clustering and recommenders (P@5 0.92).
|
More projects
| Project | What it is | Stack |
|---|---|---|
| Deep Isolation Forest | Paper implementation for unsupervised credit-card fraud detection — ROC-AUC 0.94 | PyTorch, scikit-learn |
| Fleet Management System | Multi-robot fleet manager with Dijkstra routing, lane locking and collision avoidance | Python, Tkinter |
| FlatHouse | ClickHouse → CSV export tool with JWT auth and column selection | Java, Spring Boot |
| Gym Exercises Classifier | Video action recognition for 3 exercises | TensorFlow, InceptionV3 |
| PriceForecaster | Gold price forecasting — ARIMA/SARIMAX vs ML regressors | statsmodels, XGBoost |
| SimpleChain | Minimal hash-linked blockchain with tamper detection | Python, Flask |
Languages
ML / AI
Backend & Web
Data & Infra
I'm actively looking for opportunities in Data Science, Machine Learning and AI Engineering. If my work looks like a fit for your team, I'd love to talk.
