CortexLab

Inspiration

Research is the engine of human progress — but the process of conducting research is painfully slow. Researchers spend months on literature reviews, struggle to identify novel angles in crowded fields, and burn countless hours drafting papers that follow rigid academic conventions.

We asked ourselves: What if AI could be the research assistant that every scientist deserves?

We were inspired by watching brilliant researchers get bogged down in the tedious parts of their work — not the exciting experiments or breakthrough insights, but the repetitive grind of finding papers, synthesizing trends, and formatting manuscripts. With the explosion of LLM capabilities and agentic AI systems, we saw an opportunity to build something transformative: an AI that doesn't just answer questions, but actively accelerates the entire research lifecycle from idea to publication.

What It Does

CortexLab is an AI-powered research assistant that guides you from a rough research idea to a publication-ready paper.

Here's how it works:

Describe Your Interest
Tell CortexLab what you want to research (e.g., "transformer architectures for medical imaging").
Gap Discovery
Our AI agents scan the research landscape, analyze trends, and identify untapped research opportunities that others have missed. You get a curated list of promising directions with supporting evidence.
Deep Dive & Experiment Design
Pick a direction, and CortexLab conducts a comprehensive literature review. It then generates a detailed experiment plan: baselines to compare against, datasets to use, evaluation metrics, and implementation guidance.
Paper Drafting
Upload your experiment results (logs, metrics, notes), and CortexLab drafts a complete academic paper in IMRaD format (Introduction, Methods, Results, and Discussion). Export to Word and submit to your target venue.

How We Built It

Architecture

Frontend: React 19, TypeScript, Vite, Tailwind CSS v4, Framer Motion
Backend: FastAPI (Python) with async SQLAlchemy
AI Engine: LangGraph for multi-agent orchestration, LangChain for LLM integrations
LLM Providers: Google Gemini and Groq for fast inference
Database: PostgreSQL (async) for production, SQLite for development
Auth: Google OAuth 2.0 with secure session management

The Agent System

We built a sophisticated multi-agent pipeline using LangGraph with specialized nodes:

LiteratureScout — Searches and retrieves relevant papers
TrendSynthesizer — Identifies emerging themes and patterns
GapMiner — Discovers underexplored research opportunities
DirectionGenerator — Proposes concrete research directions
DeepDiveScout — Conducts thorough literature analysis
ExperimentDesigner — Creates actionable experiment plans
PaperWriter & PaperEditor — Drafts and refines academic manuscripts

Each agent is a specialized node in our LangGraph workflow, with shared state passed between them to build comprehensive research artifacts.

Challenges We Ran Into

Agent Orchestration Complexity
Coordinating multiple AI agents that share context and build on each other required careful state design and strict information boundaries.
LLM Output Reliability
Producing consistent, structured JSON outputs required extensive prompt engineering and a JSON repair layer for edge cases.
Real-Time Streaming
We implemented Server-Sent Events (SSE) to stream agent progress so users see live updates instead of long loading screens.
Tailwind CSS v4 Migration
Major changes from v3 required multiple layout and responsiveness iterations.
Balancing Depth vs. Speed
We optimized prompts and parallelized agent execution to keep outputs thorough without excessive wait times.

Accomplishments We're Proud Of

End-to-End Research Pipeline
From “I’m curious about X” to “Here’s a draft paper about X.”
Beautiful, Polished UI
Clean white theme, smooth animations, and intuitive workflows.
Multi-Agent Architecture
A production-grade LangGraph system with 9 specialized agents working seamlessly.
Export to Word
Generate .docx files ready for academic submission.
Google OAuth Integration
Secure, production-ready authentication and session management.

What We Learned

LangGraph Is Powerful
Graph-based orchestration is ideal for complex, multi-step AI workflows.
Prompt Engineering Is Everything
Small prompt changes had massive impact on research quality.
UI/UX Matters for AI Products
A great AI needs an equally great interface.
Async Python Is Essential
Required for responsive, scalable agent execution.
Iterative Development Wins
MVP-first, refine later proved far more effective than over-engineering.