I build AI systems that actually work in production — from fine-tuning LLMs and RAG pipelines to churn prediction models running on 1M+ records. I care about clean code, reproducible experiments, and shipping things that make a measurable difference.
- 🔭 Currently building with PyTorch, Browser-use, LangGraph
- 🧠 Deep in LLMs, RAG pipelines, Multi-Agent AI
- 💬 Talk to me about AI Agents · MLOps · Data Science · Startups
- ⚡ I don't care how many times I fail. I try again.
AI / ML
PyTorch TensorFlow Scikit-learn XGBoost Hugging Face LangChain LangGraph PEFT / LoRA ChromaDB Pandas NumPy OpenCV CUDA
Languages
Python SQL JavaScript C++ Scala Bash
Backend & APIs
FastAPI Flask Django Node.js NestJS Next.js
Frontend
React Tailwind CSS HTML/CSS D3.js
Databases
PostgreSQL MySQL MongoDB Redis SQLite Cassandra
Cloud & DevOps
AWS GCP Azure Docker Kubernetes Git Grafana Prometheus
Data & Streaming
Apache Kafka Apache Hadoop Spark Kibana
Tools
Postman Figma Tableau Weights & Biases Selenium
| Project | Description | Stack |
|---|---|---|
| llm-finetune-eval | Fine-tune Mistral-7B with LoRA — 90% param reduction, tracked via W&B | PyTorch · PEFT · HuggingFace |
| semantic-search-rag | PDF Q&A with full RAG pipeline + streaming UI | LangChain · ChromaDB · Gemini |
| secureship | Dual-agent vulnerability triage & patch generation | Claude API · FastAPI · LangGraph |
| llm-inference-benchmark | GPU benchmarking: latency, throughput, memory across models | PyTorch · CUDA · Grafana · Docker |
| nerf-scene-reconstruction | NeRF pipeline — volumetric rendering + novel view synthesis | PyTorch · CUDA · OpenCV |
| multi-agent-research-assistant | 3-agent retrieval → synthesis → fact-check pipeline | LangGraph · Claude API · ChromaDB |
| churn-prediction-segmentation | Churn model on 1M+ records, +30% retention targeting lift | XGBoost · Scikit-learn · SQL · Tableau |