Pista Backend

Flask-based API backend for Pista, providing LLM-powered spreadsheet analysis and validation.

Setup

# Create and activate virtual environment with uv
uv venv
source .venv/bin/activate

# Install dependencies
uv sync

# Set environment variables (.env or export)
export ANTHROPIC_API_KEY=your_key
export OPENAI_API_KEY=your_key
export MISTRAL_API_KEY=your_key
export GEMINI_API_KEY=your_key

Running

# Development
python app/server.py

# Production (gunicorn)
gunicorn -c gunicorn.conf.py app.server:app

The server runs on http://localhost:5000 by default.

Project Structure

• server.py — Flask API with endpoints for segment generation, editing, Q&A, rubric scaffolding/verification, and chat
• dspy_programs.py — DSPy signatures and programs for LLM operations with support for Anthropic, OpenAI, Mistral, and Google
• utils.py — LLM dispatch layer, thread-safe context management, and wire format handling
• js_validator.py — JavaScript syntax validation and iterative mistake tracking
• params.py — Configuration, API keys, model endpoints, and constants
• stubs.py — Mock data for testing/development
• llm_logger.py — Logging utilities for LLM interactions
• gunicorn.conf.py — Production server configuration

API Endpoints

• POST /addin/generate — Generate segments from data
• POST /addin/edit — Edit existing segments
• POST /addin/ask — Ask questions about content
• POST /addin/scaffold-rubric — Create rubric structure
• POST /addin/verify-rubric — Validate rubric accuracy
• POST /addin/chat — Chat interactions