ResearchMind AI

Production-grade AI research platform with RAG, multi-provider AI, live LaTeX editor, and PDF compilation.

Stack

• Next.js 16 (App Router, Turbopack), TypeScript, Tailwind CSS
• RAG: pgvector + nomic-embed-text (Ollama) or text-embedding-3-small (OpenAI)
• AI: Ollama/Qwen (dev), Claude/Gemini/Groq (prod) — swappable via env var
• DB: PostgreSQL + pgvector via Prisma 7 + PrismaPg adapter
• Editor: Monaco Editor + server-side pdflatex
• State: Zustand, Jobs: Redis + BullMQ
• Auth: NextAuth.js (Google + GitHub)

Quick Start (Local Dev)

1. Prerequisites

# Start PostgreSQL with pgvector
docker run -d --name pgvector -e POSTGRES_PASSWORD=postgres \
  -p 5432:5432 pgvector/pgvector:pg16

# Start Redis
docker run -d --name redis -p 6379:6379 redis:7-alpine

#add your API key. 

2. Database Setup

# Create database and enable pgvector
psql -U postgres -h localhost -c "CREATE DATABASE researchmind;"
psql -U postgres -h localhost -d researchmind -c "CREATE EXTENSION IF NOT EXISTS vector;"

# Run migration
psql -U postgres -h localhost -d researchmind < prisma/migrations/0001_init/migration.sql

3. Environment

cp .env .env.local
# Edit .env.local — set NEXTAUTH_SECRET to a random 32-char string
# Set GOOGLE_CLIENT_ID/SECRET and/or GITHUB_CLIENT_ID/SECRET for auth

4. Run

npm install
npx prisma generate
npm run dev
# http://localhost:3000

Docker Compose (Full Stack)

docker-compose up -d
# Ollama runs on host — app connects via OLLAMA_BASE_URL

Environment Variables

Variable	Default	Description
AI_PROVIDER	ollama	ollama, claude, openai, groq, gemini
AI_MODEL	qwen2.5:14b	Model name for selected provider
EMBEDDING_PROVIDER	ollama	ollama or openai
EMBEDDING_MODEL	nomic-embed-text	Embedding model
EMBEDDING_DIM	768	Must match model output dimension
DATABASE_URL	postgresql://...	PostgreSQL with pgvector
REDIS_URL	redis://localhost:6379	For BullMQ background jobs
ANTHROPIC_API_KEY	—	For Claude in production
NEXTAUTH_SECRET	—	Random 32-char string (required)

RAG Architecture

Every AI call retrieves semantically relevant chunks before querying the model:

User query → embed → cosine search pgvector → top-8 chunks → inject into prompt → AI → ingest response back

Sources: uploaded PDFs, chat history, LaTeX sections, citations, hypotheses, arXiv abstracts.

Features

• AI Research Chat — streaming, RAG-augmented, source citations, write-to-paper
• LaTeX Editor — Monaco + templates (NeurIPS/ICML/IEEE/ACM/arXiv) + live compile
• PDF Preview — live iframe, auto-compiles on Cmd+S
• Knowledge Graph — React Flow, RAG dedup on node creation
• Hypothesis Tracker — evidence scoring via BullMQ + pgvector
• arXiv Feed — semantic similarity filtering, near-duplicate alerts
• Citation Manager — Semantic Scholar search, DOI/arXiv lookup, BibTeX export
• Corpus Analysis — up to 20 PDFs, cross-paper RAG Q&A
• Contradiction Detector — background scan via BullMQ
• Argument Validator — RAG evidence checking per section
• Figure Generator — NL to matplotlib + pgfplots LaTeX
• Novelty Score — Semantic Scholar ANN search

Screenshots

All screens share the same three-pane workspace: a feature-specific tool on the left, the Monaco LaTeX editor in the middle, and a live PDF preview on the right. Every edit flows through the same RAG context, so results in one panel immediately inform the others.

Experiment Lab

Run Python experiments in a sandboxed subprocess. Logs stream back over SSE, matplotlib figures are auto-captured as PNGs, and parameters are injected through a validated env channel so code is never interpolated into the runner.

Diagram Studio

Author Mermaid diagrams (flowcharts, sequence, ER, class, mindmap) with AI-assisted generation and refinement. The zoomable preview re-renders live as you edit, and diagrams export as SVG/PNG or drop into LaTeX as a \includegraphics block.

Hypothesis Tracker

Track research hypotheses with evidence scores. A BullMQ background worker re-evaluates every hypothesis against the project's RAG corpus whenever new papers, citations, or sections are ingested, so the scores stay fresh without blocking the UI.

arXiv Feed

Topic-aware paper feed powered by semantic similarity against your project's embeddings. The +Cite button resolves the paper via the arXiv Atom API, generates BibTeX, and ingests the abstract so future AI calls can reference it.

Citation Manager

Semantic Scholar search, DOI/arXiv lookup, and near-duplicate detection. Every added citation becomes a first-class RAG source — the AI writing tools will cite it automatically when relevant.

AI Sidebar

Specialised writing agents — Related Work synthesiser, Abstract Writer, Contributions summariser, Experiment Designer, and Reviewer Simulator. Each tool is grounded in the project's full RAG context and writes directly into the Monaco editor.

Submission Checker

Conference-ready validation for NeurIPS, ICML, IEEE, ACM, and arXiv. Checks page length, required sections, anonymisation, reference formatting, figure/table captions, and math-mode hygiene before you upload the PDF.

Research Replay

Scrub a timeline of every hypothesis update, citation add, and paper edit. Useful for writing a methods/limitations section, generating a project post-mortem, or recovering a version of the paper from a specific point in time.