Spandan Nagale

I build production-ready machine learning systems—especially multi-modal assistants and research copilots. I’m focused on turning research-grade ideas into reliable, deployable applications that users can interact with confidently.

In my recent work as a Remote ML developer, I implemented retrieval-augmented generation using FAISS embeddings and LLMs, then deployed the experience on Streamlit with an interactive UI. I also built a research copilot that integrates Arxiv, PubMed, Wikipedia, and WolframAlpha, using a LangChain agent for intelligent query handling.

I care deeply about practical quality: strict data leakage prevention through training/testing separation, portable pipelines exported as full sklearn.pipeline.Pipeline objects, and integrated explainability via feature importance and performance visualization. Projects like Veritas (privacy-first self-hosted RAG with vector search and knowledge graph traversal) and a CrewAI + Gemma multi-agent research API with FastAPI + SSE streaming reflect how I engineer for accuracy, speed, and maintainability.