abhimaurya-dev/cn_ass

Excel Mock Interviewer — Agentic MVP (Packaged)

This package contains a Streamlit conversational agent that conducts an 8-question Excel interview, keeps structured memory, evaluates answers, and produces a final report. The agent is built to prefer using LangGraph + an LLM (Hugging Face) when available, and falls back to a heuristic evaluator so you can run immediately without heavy ML dependencies.

What you get

• —app.py — Streamlit app (agent + memory + fallback evaluator)
• —requirements.txt — Python dependencies (install to enable full features)
• —Dockerfile — For containerized deployment (Hugging Face Spaces / any container host)
• —README.md — This file

How to run locally

1. 1.Create virtual env: python -m venv venv && source venv/bin/activate
2. 2.Install deps: pip install -r requirements.txt
3. 3.If you only want to run fallback heuristic mode, you only need streamlit.
4. 4.Run: streamlit run app.py
5. 5.Open the browser at the displayed URL (Streamlit default).

Hugging Face Spaces deployment (container)

1. 1.Build: docker build -t excel-interviewer .
2. 2.Run: docker run -p 7860:7860 excel-interviewer
3. 3.Alternatively push this repo to a Hugging Face Space with docker runtime.

Notes & Next steps

• —To enable full LLM-driven agent:
• —Configure a Hugging Face repo/model compatible with your hardware in setup_llm() or set HF_REPO secret in the environment / st.secrets.
• —Install transformers, accelerate, and related libs. Quantize models for 4GB VRAM using bitsandbytes where possible.
• —The app already stores structured memory in state['history']. You can extend this to vector DB memory for long-term recall.