AryaGPT v2 🤖

An agentic, self-updating personal AI assistant that answers questions about Arya Shah — built with LangGraph, ChromaDB, and Streamlit.

---

Features

• Agentic RAG — LangGraph agent with 8 tools: knowledge-base retrieval, web search (DuckDuckGo, Arya-scoped), GitHub activity, date lookup, document summarizer, conversation history search, elevator pitch generator, and follow-up question suggester
• Local embeddings — all-MiniLM-L6-v2 + cross-encoder reranking (no API key needed for embeddings)
• Multi-provider LLM — Groq, Together AI, or OpenAI, switchable live via admin panel
• Admin panel — password-gated, cookie-authenticated; change provider/model and it instantly applies to all live visitors
• Self-updating KB — push any file to data/ → GitHub Action rebuilds ChromaDB → Streamlit redeploys automatically
• Eval dashboard — P50/P95 latency, per-provider breakdown, per-query detail table, latency over time chart
• Streaming UI — dark-themed Streamlit app with source citations

---

Project Structure

aryagpt/
├── app.py                      # Main chat UI
├── pages/
│   └── admin.py                # Admin panel (password-gated)
├── agent/
│   ├── graph.py                # LangGraph agent
│   ├── tools.py                # 8 tool implementations
│   └── prompts.py              # System prompts
├── rag/
│   ├── ingest.py               # Ingestion pipeline
│   ├── retriever.py            # Retrieval + reranking
│   └── embeddings.py           # Local embedding model
├── admin/
│   └── config.py               # Provider/model config
├── db/
│   └── logger.py               # SQLite logger + eval metrics
├── data/                       # Drop your files here
├── chroma_db/                  # Auto-managed vector store
├── scripts/
│   └── ingest.py               # CLI ingest tool
├── .github/workflows/
│   └── update_kb.yml           # Auto-rebuild KB on data push
├── .streamlit/
│   ├── config.toml             # Theme config
│   └── secrets.toml            # API keys (DO NOT COMMIT)
└── requirements.txt

---

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Configure secrets

Edit .streamlit/secrets.toml:

ADMIN_PASSWORD = "your_secure_password"
ADMIN_USERNAME = "arya"
ADMIN_NAME     = "Arya Shah"

GROQ_API_KEY     = "gsk_..."
TOGETHER_API_KEY = "..."
OPENAI_API_KEY   = "sk-..."   # optional

3. Add your data

Drop any files into data/:

• .pdf — resume, documents
• .csv — structured Q&A
• .md / .txt — bios, project descriptions
• .json — structured data

4. Build the knowledge base

python scripts/ingest.py

5. Run the app

streamlit run app.py

---

Adding/Updating Data

Method 1 — GitHub push (recommended):

# Drop your file into data/
git add data/your_new_file.pdf
git commit -m "add updated resume"
git push

The GitHub Action runs automatically, rebuilds the ChromaDB index, commits it back, and Streamlit Cloud redeploys. Live in ~2 minutes.

Method 2 — Admin panel: Navigate to /admin → Knowledge Base tab → click Run Re-Ingest Now.

---

Deploying to Streamlit Cloud

1. Push this repo to GitHub
2. Go to share.streamlit.io → New app → select your repo, app.py
3. In Advanced settings → Secrets, paste your secrets.toml contents
4. Deploy

The GitHub Action needs a GITHUB_TOKEN secret with write access — this is provided automatically by GitHub Actions.

---

Admin Panel

Navigate to /admin (linked in the Streamlit sidebar). Login with your ADMIN_USERNAME / ADMIN_PASSWORD.

Tabs:

Tab	What you can do
⚙️ Provider & Model	Switch LLM provider/model live — applies instantly to all visitors
📚 Knowledge Base	View KB stats, list source files, trigger manual re-ingest
💬 Conversation Logs	Browse and search all conversations (SQLite)
📊 Eval Dashboard	P50/P95 latency, provider breakdown, per-query table, latency chart

---

Environment Variables

All secrets should be in .streamlit/secrets.toml (local) or Streamlit Cloud secrets (production).

Variable	Required	Description
ADMIN_PASSWORD	✅	Admin panel login password
ADMIN_USERNAME	optional	Admin panel login username
GROQ_API_KEY	✅ (if using Groq)	Groq API key
TOGETHER_API_KEY	✅ (if using Together)	Together AI API key
OPENAI_API_KEY	optional	OpenAI API key