FELIPE WINSCHFull Stack Developer
Back to projects

Legal case management · Applied AI

LexCore AI

A prototype bringing clients, cases, documents and reports together. It combines a Rust backend and React interface with PDF question answering through Ollama.

Prototype

  • Rust
  • Tuono
  • React
  • TypeScript
  • PostgreSQL
  • Ollama

My contribution

A project maintained by felpzw, with public history attributed to that profile. The reviewed scope combines React pages, Rust handlers, relational modeling and Ollama integration on top of open-source technologies.

Problem

Bring client, case and document information into one application and query PDF text without switching between separate tools.

Context and scope

Previously called Jurídico IA, the repository presents a development prototype. It includes client, case and document CRUD, reports and an AI-assisted analysis screen. Homepage promotional text is not treated as measured results.

Architecture

Tuono connects React/TypeScript pages to Rust handlers. PostgreSQL stores relationships and files in BYTEA columns; Recharts displays report aggregates.

For document analysis, the backend retrieves a PDF by filename, extracts text with pdf-extract and combines it with the question. A non-streaming Ollama request returns the answer to the interface.

  1. React / Tuono · document and question
  2. Rust / PostgreSQL · PDF retrieval and extraction
  3. Ollama · response generation from text
Implemented PDF question-answering flow, without OCR or vector search.

Decisions and trade-offs

Keeping files and metadata in the same database centralizes access but increases the data volume PostgreSQL must handle. Sending text directly in the prompt simplifies the pipeline but provides neither passage retrieval nor index-based context selection. These trade-offs are inferred from the implementation.

Tests and evidence

Static review covered documentation and the document and Ollama handlers. No automated answer-evaluation suite was found. The post_test endpoint is route code, not a test suite. The application and its services were not run during this review.

Limitations

This flow has no OCR, document chunking or vector index. The endpoint does not implement contract or petition generation. If Ollama points to a remote service, extracted text is sent to that service.

Database initialization and table-deletion endpoints lack authentication, and development configuration is tracked in Git. This is a local experimental setup; the review validates neither use with real legal documents nor answer quality.

Explore the project

Read the project overview and follow the flow between documentos.rs and ollama.rs. The setup guide covers Rust, Node, Tuono 0.19.7, PostgreSQL and model installation.

After cloning the indicated revision, configure var.env and a disposable database before starting services. Use synthetic PDFs; initialization in /configuracoes can delete tables. The commands below reflect the guide but were not run in this review.

npm ci
docker compose up -d
tuono dev

Sources checked on 2026-09-27 · Content v1 · Revision c151ed8.