ML/AI Engineer · Python · PyTorch · LangGraph · FastAPI
Designing and building machine learning systems — from model training to production APIs.
- Building ML systems with a focus on agentic AI, recommender systems, NLP/LLMs, and multimodal learning
- Honors research in EEG-based attention classification and cross-subject biomarker discovery
- Experience deploying end-to-end pipelines: data ingestion → model inference → REST API
- Interested in LangGraph agents, RAG systems, Graph Neural Networks, and signal processing
- Based in Chennai, Tamil Nadu, India
| Project | Description | Tech |
|---|---|---|
| LangGraph-ReAct-Reasoning-Agent | Tool-augmented reasoning agent using LangGraph ReAct loop with intent classification, multi-step reasoning, and a self-critique review pass. Supports Tavily search, Wikipedia lookup, and calculator tools. SSE streaming + FastAPI frontend. | Python, LangGraph, LangChain, Groq, Tavily, FastAPI |
| Arxiv-RAG-Assistant | Research assistant using hybrid search (BM25 + dense embeddings + cross-encoder reranking) to answer questions with citations from scientific papers. Powered by Llama 3 via Groq. | Python, FastAPI, BM25, ChromaDB, RAG, Llama 3 |
| Project | Description | Tech |
|---|---|---|
| FasRec-AI-Engine | Multimodal fashion recommender for 44k+ products using CLIP image embeddings, SBERT text embeddings, and FAISS HNSW ANN search. FastAPI backend with Cloudflare R2 storage. Deployed on Render. | Python, PyTorch, CLIP, SBERT, FAISS, FastAPI |
| Simple-Arxiv-Paper-Recommender | NLP prototype recommending related arXiv papers via TF-IDF and embedding-based abstract similarity | Python, NLP, TF-IDF, Embeddings |
| Simple-Fashion-Recommendation-System-Phase1 | Content-based fashion recommender using K-Means clustering on product attributes and image features | Python, K-Means, Jupyter |
| Project | Description | Tech |
|---|---|---|
| Loan-Collections-Recovery-Engine | End-to-end SQL + ML pipeline on the Home Credit dataset. Warehouse design (staging → dims → facts → analytical marts), delinquency feature engineering, recovery propensity scoring with LightGBM, and dashboard-ready Power BI exports. | Python, PostgreSQL, SQLAlchemy, LightGBM, scikit-learn, pandas |
| Project | Description | Tech |
|---|---|---|
| AttentionEEG | Attention classification across multimodal EEG stimuli (audio, video, text) | Python, MNE, Scikit-learn, Jupyter |
| Modality-Invariant Biomarker Discovery | Identifying stable EEG biomarkers for attention that generalize across audio, video, and text modalities | Python, EEG Signal Processing |
| Project | Description | Tech |
|---|---|---|
| DbPedia Entity Classification (GCN & GraphSAGE) | Multiclass text classification on DBpedia-14 using GCN and GraphSAGE architectures | Python, PyTorch Geometric, Jupyter |
| Project | Description | Tech |
|---|---|---|
| Protein Remote Homology Detection | Remote protein homolog detection via ESM-2 sequence embedding alignment and Smith-Waterman scoring | Python, ESM-2, Sequence Embeddings |
| Project | Description | Tech |
|---|---|---|
| EstateCraft | Full-stack MERN platform for property listings with Stripe payments and role-based auth | React, Redux, Node.js, MongoDB, Stripe |
| NourishNest | Nutrition and diet planning platform with recipe database and meal tracking via Spoonacular API | Node.js, Express, MongoDB, EJS |
| Smart-Scheduler | Context-aware task scheduling assistant | Python, JavaScript |
| Smart-Pg-Review-Platform | PG accommodation review and discovery platform | TypeScript |
| Project | Description | Tech |
|---|---|---|
| CSRF Attack & Defense Demo | Demonstrates CSRF vulnerability and OWASP-compliant defense mechanisms | Node.js, EJS |