System design

Open-model analytics architecture for scalable file intelligence.

The architecture separates frontend UX, API routing, AI orchestration, data processing, storage, and model provider fallback so each layer can evolve independently.

Frontend

Next.js 14, Tailwind, shadcn/ui, chat UI, upload panel, chart renderer, export trigger, workspace management.

API Gateway

FastAPI routes, JWT auth, rate limiting, request validation, upload endpoints, and SSE streaming.

AI orchestration

LangChain and LlamaIndex for prompts, RAG, model routing, tool use, memory, and response shaping.

Processing engine

pandas, numpy, pdfplumber, PyMuPDF, sqlparse, pytesseract, OpenCV, Plotly, and export libraries.

Storage

Supabase PostgreSQL, object storage, file metadata, user workspaces, generated charts, and chat history.

Model fallback

Groq primary, OpenRouter fallback, Hugging Face inference, and Ollama/local enterprise mode.

Data flow

File analysis request

User uploads a file through the product frontend.
FastAPI validates type and size, stores the file, and returns a file ID.
Background processing profiles the file, extracts text/tables, and builds data summaries.
Embeddings are stored under the user and project namespace for retrieval.
User asks a question; the API builds context from the prompt, file metadata, and retrieved chunks.
Model provider returns streamed response; chart JSON is generated if needed.
Frontend renders text, charts, confidence states, and export actions.

Tech stack

Recommended implementation choices

Layer	Choice	Reason
Frontend	Next.js 14	Fast React UI, SSR/SSG, and easy Vercel deployment.
Backend	FastAPI	Native Python data library integration and async APIs.
Charts	Plotly + Recharts	Interactive analytics with export-friendly chart output.
Vector DB	ChromaDB to Qdrant	Start light in MVP, migrate for production scale.
Database	Supabase PostgreSQL	Managed DB, RLS, auth, and instant APIs.
Export	WeasyPrint + python-docx + python-pptx	Server-side professional report generation.