Graph + Vectors + LLM: Building an Adaptive ENEM Tutor with Neo4j RAG and Gemini

Learn how to build an ENEM tutor using Neo4j RAG and Gemini, combining a knowledge graph with vector search and AI for adaptive quizzes and essay correction.

Google Gemini API Gemini Embeddings Neo4j FastAPI Redis

Overview

Griô is a free AI-powered learning platform for Brazil’s national university
entrance exam (ENEM). The core is a knowledge graph (Neo4j) modeling the
entire official ENEM competency matrix — areas, competencies, skills, topics,
subtopics — with real exam questions as nodes, each carrying a 768-dimensional
Gemini embedding.

When a student asks for a quiz, the system runs asymmetric semantic search
(RETRIEVAL_QUERY vs RETRIEVAL_DOCUMENT tasks) against the question embeddings
using cosine similarity, then cross-references the graph to exclude questions
the student has already answered via traversal of HAS_PROFICIENCY and ANSWERED
relationships. If vector search returns fewer results than needed, a taxonomic
graph traversal fallback kicks in automatically.

The platform also includes a manuscript essay correction module: the student
photographs a handwritten essay, Gemini Vision transcribes it verbatim
(preserving grammar errors), the student confirms the transcription, and a
structured JSON evaluation scores all 5 official ENEM rubric competencies
(0–200 each) using chain-of-thought prompting with reason-before-score
ordering and self-consistency sampling.

Live demo: working system, real graph queries, RAG pipeline, essay evaluation.

Video

Links

https://grio.ricaeldurand.com
AI-driven platform generating personalized study plans and explanations targeting weaknesses.

Tech stack

Google Gemini API

Integrate Google's multimodal Gemini models (Pro, Flash) into your application via REST or SDKs: generate content, process up to 1000-page PDFs, and execute code with a 2-million token context.

The Gemini API delivers Google's most advanced models (Gemini 3 Pro, 2.5 Flash) directly into your applications. Leverage its multimodal power: process text, images, video, and audio inputs for tasks like content generation, summarization, and visual understanding. Utilize key features including the 2-million token context window, structured JSON output, and function calling to build complex, reliable agents. Choose your integration: use the standard REST API, streaming endpoints (SSE), or the Live API (WebSockets) for real-time conversational experiences. Get started with the free tier and robust SDKs for Python, Java, Go, and more.

https://ai.google.dev/

View projects
Gemini Embeddings

Google's high-performance vector representation models designed to map text, images, video, audio, and documents into a single, unified embedding space.

Gemini Embeddings convert complex, multi-format inputs into high-dimensional numerical vectors to power semantic search, classification, and retrieval-augmented generation (RAG) systems. With the release of Gemini Embedding 2, developers can bypass traditional, multi-step processing pipelines (such as transcribing audio before embedding) and directly map diverse media types into a single 3072-dimensional vector space. Available through the Gemini API and Vertex AI, this technology simplifies cross-modal search and delivers context-aware similarity matching across more than 100 languages.

https://ai.google.dev/gemini-api/docs/embeddings

View projects
Neo4j

Neo4j is the world's leading native graph database, purpose-built for high-performance management and traversal of connected data.

Neo4j is the leading native graph database, leveraging the Property Graph Model (nodes, relationships, properties) for data storage and retrieval . It is an ACID-compliant, high-performance platform designed for managing highly connected data at scale (billions of nodes) . Queries execute using the declarative Cypher query language, which simplifies complex traversals that would cripple relational systems with numerous JOINs . This architecture delivers orders of magnitude performance improvements (often minutes to milliseconds) for relationship-based queries . Major use cases include fraud detection, recommendation engines, and knowledge graphs for AI, trusted by 84 of the Fortune 100 .

https://neo4j.com

View projects
FastAPI

FastAPI is a modern, high-performance Python web framework for building APIs with automatic OpenAPI documentation.

FastAPI is a robust, high-speed Python web framework: it is built on Starlette (for async capabilities) and Pydantic (for data validation and serialization). Leveraging standard Python 3.8+ type hints, the framework automatically generates interactive API documentation (Swagger UI/ReDoc) and enforces data validation, effectively reducing developer-induced errors by an estimated 40%. This architecture delivers performance on par with Node.js and Go, significantly increasing feature development speed (up to 300% faster). It is production-ready, fully supporting OpenAPI and JSON Schema standards for all API specifications.

https://fastapi.tiangolo.com

View projects
Redis

Redis is the ultra-fast, open-source, in-memory data structure store: a powerful NoSQL key/value database.

This is your go-to for low-latency data operations. Redis operates primarily in memory, delivering sub-millisecond response times for real-time applications (think: session storage, leaderboards, and caching). It functions as more than just a key/value store; it’s a versatile data structure server supporting Strings, Hashes, Lists, Sets, Sorted Sets, and JSON. Leverage its Pub/Sub capabilities for message brokering, or rely on its optional persistence for durability. Deploy it for high-speed caching to offload your primary database, or use it as a primary database for high-throughput microservices.

https://redis.io

View projects

Graph + Vectors + LLM: Building an Adaptive ENEM Tutor with Neo4j RAG and Gemini

Related talks