RAG Platform

Enterprise-Grade RAG Platform

AI-Powered Research Workbench for Academic Literature

Upload your PDFs and chat with your literature using state-of-the-art retrieval and citation-grounded responses. Built for researchers and evidence synthesis teams.

How It Works

1. Upload Documents

Upload PDFs, research papers, reports, or any documents you want to make searchable.

2. AI Processing

Documents are automatically parsed, chunked, and indexed for intelligent retrieval.

3. Chat & Search

Ask questions in natural language and get AI-synthesized answers with citations.

Core Features

Powerful Capabilities

Built with state-of-the-art AI technology for accurate, citation-grounded responses

Smart Document Ingestion

MinerU GPU parsing for tables, figures & equations
Page-aware chunking with 256-word segments
CrossRef/OpenAlex metadata enrichment
Zotero collection sync & bibliography matching

Hybrid Search

Dense vectors with Cohere embed-v4 (1536d)
Sparse BM25-style keyword matching
HyDE hypothetical document expansion
Cohere rerank-v3.5 precision filtering

Citation-Grounded Synthesis

ALCE-style Claim Ledger two-pass extraction
Inline citations [Author, Year, p.X]
Click-to-navigate PDF viewer integration
MiniCheck NLI post-hoc verification

Study Design Classification

Auto-classify 22 study types (RCT, DiD, RDD, etc.)
Filter evidence by study design
Contextual retrieval prefixes (49% fewer failures)
Parent-child chunking for full context

Multi-Model Support

Per-RAG model selection (GPT-4o, Claude, Gemini)
Custom system prompts per project
Prompt versioning (up to 10 versions)
OpenRouter unified LLM access

Admin & Observability

Full admin dashboard with usage stats
Cost tracking per model & project
Langfuse LLM tracing & debugging
Quality metrics & baseline experiments

Quality Assurance

Rigorous Citation Quality

Evaluated using MiniCheck NLI verification on golden queries

66.5%

Faithfulness

Claims grounded in retrieved documents

98.9%

Citation Recall

All claims properly cited

98.9%

Citation Precision

Citations point to correct sources

Technology

Enterprise-Grade Stack

Frontend

Next.js 15, TypeScript, Tailwind CSS

Backend

FastAPI, Python 3.12, Pydantic v2

Auth & Database

Supabase (JWT + RLS), PostgreSQL

Vector Search

Qdrant Cloud (dense + sparse vectors)

LLM & Embeddings

OpenRouter, Cohere embed-v4 & rerank-v3.5

PDF Processing

MinerU 2.6.8 on Modal (T4 GPU)

Use Cases

Research Teams

Upload literature reviews, papers, and data. Team members can quickly find relevant information across all documents.

Knowledge Management