Employee Handbook AI.

01

Context & problem

The prototype explores answering employee-policy questions from a handbook. It uses a self-built synthetic handbook, so the work can demonstrate a retrieval workflow without presenting real company policy.

02

My role

I took the project from use-case definition to a deployed Streamlit prototype. My work covered the test knowledge base, technical workflow, retrieval layer, API integration, and interface.

03

How it works

Documents are split into chunks and embedded with Sentence Transformers. Vector search retrieves relevant passages before the Groq LLM API generates an answer.

The Streamlit interface connects the question, retrieval, and generation steps, with source references to help readers trace the response.

04

Key decisions

I used retrieval to supply document context to the language model and included source references in the answer flow.

Using synthetic handbook content kept the prototype focused on the technical workflow. Source references support traceability; they do not guarantee a correct answer.

05

What I built

Built document chunking, embeddings, semantic search, API integration, and a Streamlit interface. Deployed the proof of concept to Streamlit Cloud.

Iterated on prompts within the prototype. No formal answer-accuracy score or user-impact improvement is claimed.

06

Limits & next steps

The system has not been formally tested with real users. A working demo does not establish reliability for live HR policy decisions.

Evaluating retrieval relevance and answer faithfulness on a defined question set would be a meaningful next validation step; no such evaluation result is claimed here.