An HR policy assistant on retrieval-augmented generation
A retrieval-augmented chatbot that answers HR policy questions from the policy documents themselves.
Measured outcomes
~60%
faster HR policy query resolution
Shipped as an internal tool: FAISS retrieval feeds Llama inference, replacing manual lookups and keyword search.
Context
Answering an HR policy question meant someone looking it up by hand in the policy documents. At Parkar Global Technologies I built a chatbot that answers those questions from the documents themselves, using retrieval-augmented generation.
Architecture
Pick one to light up the parts behind it.
HR policy assistant · Python
- FastAPI · API service
- LangChain · orchestration
- Llama · inference
- FAISS · semantic retrieval
- Question · HR policy
- HR policies · source documents
Connections
- Question to FastAPI
- FastAPI to LangChain
- LangChain to Llama
- LangChain to FAISS
- HR policies to FAISS
A question comes in through a FastAPI service, and LangChain coordinates the rest. FAISS semantic search finds the parts of the HR policy documents that matter for the question, and Llama inference writes the answer with them as context. It's written in Python.
What I built
My part
- The assistant itself, on a retrieval-augmented generation design.
- Semantic document retrieval with FAISS, and the Llama inference that answers in context.
- The Python service around them, with FastAPI and LangChain.
Results
I shipped the chatbot as an internal tool, and it replaced manual lookups and keyword searches through the policy documents. The figure at the top of this page is how much faster that made resolving HR policy queries.