Semantic Search Studio
An LLM-powered semantic search engine over a private document knowledge base, with cited answers.
A user uploads documents (PDF, Markdown, or plain text), and the system makes that corpus searchable by meaning rather than by keyword. When the user asks a question in natural language, the engine retrieves the most semantically relevant passages and uses Claude to synthesize a single, grounded answer with inline citations back to the source material.
Try it live → — running right here on this site: upload a document (PDF, Markdown, or text) and ask a question about it. (Answer synthesis calls Claude and needs an API key configured on the server; document upload and passage retrieval work regardless.)
The problem it solves
Traditional keyword search (Ctrl+F, BM25, SQL LIKE) fails when the
user's wording doesn't match the document's wording — someone searching "how
do I get my money back" won't find a paragraph titled "Refund Policy."
Semantic search closes that gap by matching on intent. Layering an LLM on
top turns a list of blue links into a direct, cited answer: the difference
between finding and knowing.
Tech stack
| Layer | Choice |
|---|---|
| Backend | Python 3.11+ · FastAPI · Uvicorn · ChromaDB · Anthropic SDK · Sentence Transformers · pypdf · Pydantic |
| Frontend | React 18 · Vite |
| Containerization | Docker |
The live demo above runs this exact backend; to see the ingestion, retrieval, and generation pipeline, view the source.