How simpnick6703/talentscout-hiring-ai-assistant works
Project type and purpose
• This repository implements a full-stack conversational AI assistant (“TalentScout”) for conducting structured interviews, plus a separate “Interview Reader” tool for visualizing exported interviews.
• The main app is built with Streamlit (app.py) and Python modules that orchestrate LLM calls, data processing, and UI rendering. The interview reader is a lightweight Flask/Jinja app under interview_reader/.
File structure
• Root directory:
• app.py – Streamlit entry point and route definitions
• chatbot.py – Core interview flow logic and LLM prompt orchestration
• config.py – LLM provider settings, UI theme flags, question-flow parameters
• data_handler.py – JSON serialization, file I/O, export logic
• utils.py – Helper functions (e.g., timestamp formatting)
• static/styles.css – Global CSS for any custom UI theming
• requirements.txt – Python dependencies
• .github/ – CI workflows and Dependabot settings
• interview_reader/ sub-app:
• simple_reader.py – Flask application to serve JSON-backed interview views
• templates/ – Jinja2 HTML templates (index, upload, conversation, view_interview)
• README.md and requirements.txt for the reader tool
Key insights from the README
• Backend: Python + Streamlit
• Frontend: Streamlit UI + static CSS
• AI Integration: OpenAI Python SDK (or LM Studio)
• Data store: File-based JSON exports
• Interview phases: greeting → data gathering → technical questions → wrap-up
• Interview Reader: separate app for browsing and analyzing JSON exports
a. Main components to represent:
• Candidate’s browser or user interface
• Streamlit server (app.py)
• Chatbot engine (chatbot.py)
• Configuration manager (config.py)
• Data handler and JSON storage (data_handler.py + file system)
• External LLM provider (OpenAI / LM Studio API)
• Interview Reader service (simple_reader.py + templates)
b. Relationships and data flow:
1. Browser ↔ Streamlit server: UI events and responses
2. Streamlit server → Chatbot engine: “next question” requests
3. Chatbot engine → LLM provider: API calls for message generation
4. Chatbot engine → Data handler: append conversation chunks, export JSON
5. Browser → Interview Reader: JSON upload or file browse
6. Interview Reader → Templates: render conversation history
c. Architectural patterns and design principles:
– Separation of concerns: UI layer (Streamlit) vs. business logic (chatbot.py) vs. persistence (data_handler.py)
– Modular configuration (config.py) decouples environment and model settings
– MVC-style pattern in interview_reader (Flask routes, templates, static assets)
– Provider-agnostic LLM integration via adapter in config/chatbot
d. Relevant technologies and frameworks:
– Streamlit for UI
– Flask + Jinja2 for reader tool
– OpenAI Python SDK (or compatible LM Studio client)
– JSON file system for interview persistence