Interactive Conversational Intelligence Dashboard

Ask questions about your data in plain English. The app translates them into SQL using Google Gemini, runs the query, and renders a chart automatically.

---

Project Structure

├── app.py       # Streamlit frontend
├── main.py      # FastAPI backend
├── engine.py    # DB manager, session manager, LLM interface
└── models.py    # Pydantic models

---

Dependencies

pip install fastapi uvicorn streamlit requests pandas plotly google-generativeai python-multipart

PostgreSQL support (optional):

pip install psycopg2-binary

---

How to Run

streamlit run app.py

On first launch, a startup screen asks for your Gemini API key. Enter it and click Start backend — the app spawns the FastAPI server automatically and loads once it is ready.

To skip the prompt, set the key as an environment variable beforehand:

export GEMINI_API_KEY=AIza...

Get a key at https://aistudio.google.com/app/apikey

---

How It Works

Every question goes through this pipeline:

User question
     │
     ▼
Validate session → Extract schema → Call Gemini (question + schema + history)
     │
     ▼
Gemini returns SQL + chart config (JSON)
     │
     ▼
Validate SQL safety → Execute query → Return data to frontend
     │
     ▼
Plotly renders the chart

Key internals:

• CSV ingestion — uploaded files are loaded into an in-memory SQLite database via pandas. The filename becomes the table name.
• LLM call — the database schema and last 20 turns of conversation are sent to gemini-flash with each request, enabling follow-up questions.
• SQL safety — before execution, all SQL is checked against a blocklist (DROP, DELETE, INSERT, UPDATE, etc.). Only SELECT statements are allowed.
• Chart selection — Gemini picks the chart type based on the data shape (bar, line, pie, scatter, area, or table).
• Backend auto-launch — the Streamlit app spawns main.py as a child process with the API key injected into its environment. The process is terminated automatically when Streamlit exits.

---

Supported Data Sources

Source	How to connect
CSV	Upload via the sidebar — ingested into in-memory SQLite
SQLite	Provide the file path
PostgreSQL	Provide a DSN: postgresql://user:pass@host:5432/db

---

Known Limitations

• Dates stored as strings (e.g. 13-04-2022) can cause incorrect results for date range queries.
• Complex multi-part questions often fail — break them into smaller focused questions.
• CSV sessions are lost if the backend restarts and the file must be re-uploaded.