AI Agent Project

Voice-first chatbot stack with real-time dashboard, long-term memory, benchmarks, and PDF tooling.

What's Inside

• chatbot/ – voice chatbot pipeline (Whisper STT, NeuTTS/Piper TTS, emotion/personality analysis, MemoBase memory) plus Flask REST/SSE backend.
• frontend/ – React + Vite dashboard that streams live metrics from the backend.
• memobase/ – embedded MemoBase server plus MCP server and Go/TS clients.
• chatbot/tests/ – regression and benchmark suites (conversation extraction, prompt sweeps, Ollama cache tests, Whisper accuracy).
• chatbot/benchmark/ – LLM-as-judge benchmark and 100-sample dataset for memory citation.
• agentic-report-gen/ – MinerU-based PDF → Markdown converter (GPU-first, ARM64 friendly).
• neutts/ – on-device TTS models and runnable examples.

Run the Stack

Prereqs: Python 3.10+, Node 18+, Docker, Ollama (for local LLMs).

./start_all.sh  # starts MemoBase (docker), chatbot API, and frontend
# Dashboard: http://localhost:3000  (API on :5000, MemoBase on :8019)

Manual entry points:

cd chatbot && python -m app.server   # REST/SSE backend
cd chatbot && python -m app.app      # interactive voice chatbot
cd frontend && npm install && npm run dev

Ports

• 3000: Frontend dashboard
• 5000: Chatbot API
• 8019: MemoBase
• 11434: Ollama (LLM runtime)