Inspira

Inspira is a multi-modal RAG assistant that helps users analyze uploaded files (PDF, PPT, images, text) and generate insights through a LangGraph reasoning workflow.

Frontend demo: https://inspira.innospace.dev/ (to save token cost, only frontend experience is currently enabled).

Features

• Multi-modal ingestion (documents + images)
• LangGraph reasoning pipeline (memory_retrieve -> classify_intent -> tool_router/refine -> generate_response)
• FastAPI backend with sync and streaming chat endpoints
• React + Vite frontend workspace experience
• Offline RAGAS evaluation pipeline with dataset cleaning utilities

Tech Stack

• Backend: Python, FastAPI, LangGraph, LangChain, OpenAI API, SQLModel
• Frontend: React, TypeScript, Vite, Tailwind CSS
• Evaluation: RAGAS, HuggingFace datasets

AWS Infrastructure & Services

• Access & Delivery: Static assets are hosted on Amazon S3 and delivered via CloudFront to ensure global low-latency access.
• Compute: Backend services are containerized using Docker and deployed on Amazon ECS (Elastic Container Service). EC2 GPU-optimized instances are selected for multimodal processing, as the large memory footprint of models like CLIP makes serverless (AWS Lambda) inefficient for this specific use case.
• Data & Memory: Amazon RDS manages structured user data, while Amazon ElastiCache handles high-frequency memory tasks. User-uploaded inspiration elements are stored in S3 for durability and cost-efficiency.
• Asynchronous Processing: Long-running tasks, such as embedding a 100-page technical paper or a high-resolution image gallery, are managed via Amazon SQS queues to prevent request timeouts and ensure a smooth user journey.

Repository Structure

Inspira/
├── backend/
│   ├── main.py                      # FastAPI app entry
│   ├── routers/                     # REST routes (stacks/files/ai)
│   ├── reasoning/                   # LangGraph state, nodes, tools
│   ├── file_processor/              # PDF/PPT/image/text processing
│   ├── rag_engine/                  # Retrieval and embedding logic
│   ├── evaluation/ragas/            # RAGAS scripts and reports
│   └── requirements.txt
├── frontend/
│   ├── src/
│   └── package.json
└── README.md

Prerequisites

• Python 3.10+
• Node.js 18+
• Conda or venv (recommended)
• OpenAI API key

Backend Setup

1. Create and activate environment

conda create -n inspira python=3.10 -y
conda activate inspira

2. Install dependencies

cd backend
pip install -r requirements.txt

3. Create backend/.env

Required minimum:

OPENAI_API_KEY=your_key_here
OPENAI_CHAT_MODEL=gpt-4o-mini
OPENAI_VISION_MODEL=gpt-4o-mini

Optional commonly used settings:

APP_MODE=local
REDIS_URL=redis://localhost:6379/0
RETRIEVAL_CACHE_ENABLED=true
MEMORY_ENABLED=true
RAGAS_LOG_CHAT_SAMPLES=1

4. Run backend (from repository root)

uvicorn backend.main:app --reload

Health check:

GET http://127.0.0.1:8000/health

Frontend Setup

cd frontend
npm install
npm run dev

Default dev URL is usually http://127.0.0.1:5173.

API Quick Reference

• POST /chat — non-streaming chat
• POST /chat/stream — SSE streaming chat
• GET /stacks / POST /stacks — stack management
• POST /stacks/{stack_id}/files — file upload

License

MIT