🔹 my goto stack

• Languages & Frameworks — Python (FastAPI) · C++ · Go · JavaScript · TypeScript (Node.js)
• Databases & Caching — PostgreSQL · MySQL · MongoDB · Redis · ChromaDB · Qdrant (Vector DBs)
• Infrastructure & DevOps — Docker · AWS (EC2/S3) · Git · CI/CD · MCP (Model Context Protocol)
• Concepts — RAG · Hybrid Search · Low-Level Design (LLD) · System Design · Agentic Workflows · Neural Networks · Supervised/Unsupervised Learning · Fine-Tuning · Feature Engineering · Evaluation Metrics
• AI/ML Frameworks — PyTorch · TensorFlow · LangChain (AI Agents) · LangGraph · Scikit-learn · Deep Learning · NLP

🔹 current work

• ResearchFlow — Multi-agent LangGraph research pipeline with Streamlit + FastAPI, Mistral AI & Tavily. 0.94 relevance score.
• AI Recruiting Matching Agent — LangGraph + ChromaDB/BM25 hybrid retrieval with a real eval harness. 0.92 recall@3, 0.90 nDCG@3.
• AI Video Assistant (RAG) — Multimodal transcription and retrieval using Whisper + Mistral AI.

🔹 open source

• Contributor to mem0 — added a GoogleGenAIEmbedding class, reviewed a native Mistral LLM provider PR.
• Merged PR in ragflow , repowise.

🔹 building from zero

• I build everything from scratch — no shortcuts, no boilerplate templates I don't understand. Every project starts as a blank file and gets shaped by real problems, real edge cases, and real eval numbers, not by copying what already exists. But the point was never just to build for myself — it's to build things people actually use and feel better for having used. If a project doesn't make someone's work easier or their day a little less painful, I don't consider it done.