Featured Project — LLM Application
Portfolio RAG Chatbot
A chatbot that answers questions about my work — projects, skills, experience — grounded in a curated knowledge base of my portfolio instead of a language model’s imagination.
The Problem
Recruiters and clients ask the same questions repeatedly: What has he built? Which models? Is he available? A static page makes them hunt; a grounded chatbot answers in seconds — and because it retrieves from my actual portfolio content, it does not invent projects or skills I do not have.
What I Built
- A retrieval pipeline over a hand-maintained knowledge base describing every project, role and skill.
- Retrieval tuned per project: one strong chunk per project plus header-keyword boosting, so questions land on the right project instead of a blend of three.
- A chat interface served as a public Streamlit app.
Technical Approach
- Embeddings: MiniLM sentence embeddings; index: FAISS.
- Generation: an OpenRouter-hosted LLM prompted with the retrieved passages and instructed to answer only from them.
- Iteration: free-tier models were tested and rejected when they leaked chain-of-thought or failed; the current model was selected for clean, grounded answers.
Technologies & Stack
Results
Live and public — ask it about any project on this site and it answers from the same facts, with the project it drew from.
Limitations
The knowledge base is updated manually, so brand-new work appears only after an update. Free LLM tiers can occasionally rate-limit under load.