Curie

Bioinformatics AI Platform

Sept 2025 - Present

Full-stack bioinformatics platform with AI assistance for biomedical research. Features real-time chat with LLM integration, RAG system with PubMed database, user authentication, and WebSocket connections for live query processing. Designed for researchers to analyze and connect their data to biomedical literature with AI support.

Stack

The Problem

Modern scientific research faces an exponential growth of publications and structured data, making literature review and knowledge extraction a slow and error-prone task. Researchers lose valuable hours navigating through thousands of PDF articles and scattered databases to find specific answers, significantly delaying progress in biotechnological and medical projects.

The Solution

Curie is an intelligent analysis platform that uses RAG (Retrieval-Augmented Generation) architectures to transform how scientists interact with information. By allowing the upload of local documents (PDF, DOCX), Curie relates data with scientific literature and integrates databases like arXiv and PubMed, acting as a domain expert that can answer complex questions accurately, citing direct sources and synthesizing technical information immediately.

Architecture

Frontend

Built with Next.js, React, and TypeScript, it provides an intuitive and dynamic chat environment. It uses Tailwind CSS for a modern and functional aesthetic, optimized for both file management and the visualization of detailed scientific results using Lucide icons.

Backend

Implemented with FastAPI and Python, the backend orchestrates the RAG engine. It uses SQLAlchemy with PostgreSQL (Supabase) for data persistence and pgvector for vector similarity search, allowing the retrieval of the exact fragment of information needed for each query.

Model Integration

The system communicates with leading LLM providers such as OpenAI and OpenRouter (leveraging models like Gemini or Claude), ensuring that answers are contextual, precise, and adjusted to the required scientific rigor.

Processing & Analysis Flow

When a user uploads a document, it goes through a process of segmentation (chunking) and embedding generation (OpenAI API) before being indexed. During a query, the engine searches for the most relevant fragments in the vector database and injects them into the language model's context. The result is a coherent response that not only resolves the researcher's doubt but also maintains the traceability of the original data, allowing for constant scientific validation.

Deployment & Scalability

Curie is fully containerized using Docker, ensuring a consistent environment from development to production. Deployment is targeted toward modern infrastructures like Fly.io (for the backend with structured logs support) and Vercel (for the frontend), allowing for efficient scaling and simplified maintenance that favors the integration of new data sources and improvements in processing models.

Features

Natural language

Most research tools force you to learn their logic. Curie works the other way around — you write exactly as you think, in plain language, and it understands what you're looking for. No special syntax, no structured queries, no learning curve.

Search databases

Instead of opening five tabs and manually cross-referencing results, Curie connects directly to PubMed, Semantic Scholar, arXiv, Web of Science, and Google Scholar. You ask once, and it searches across all of them at the same time, filtering down to what's actually relevant to your question.

Import files

Academic papers are long. Curie handles that — you can upload PDFs, DOCXs, TXTs, and MDs of up to 250 pages and it will process the entire document, not just skim the abstract. That means you can ask specific questions about methodology, results, or references buried deep in the file and actually get accurate answers.

Contextual answers

A single question rarely gets you where you need to go. With Curie, every follow-up message builds on everything said before — it remembers the full conversation, so you can refine your question, challenge an answer, or go deeper into a topic without having to repeat yourself or start over.

Code, schemas and more

When words aren't enough, Curie generates. Visual diagrams, data schemas, code snippets, formatted tables — anything you might need to take an idea further or present it clearly, ready to copy and use directly.