Civis RJ - Predictive Command Center

An AI-powered command center that automates public works auditing, predicts delays, and stops the waste of public funds.

Comment

Inspiration

Anyone who tracks public spending knows the frustration of seeing construction projects delayed, abandoned, or heavily over budget. Manual auditing of government contracts takes weeks of technical analysis and is super vulnerable to human error or omitted clauses. We built Civis RJ because we realized that while cities are growing fast, the oversight of public funds is stuck in the past. We wanted to build a tool that actually predicts risks before the money is wasted.

What it does

Civis RJ is a predictive command center that acts as an AI auditor for public works. Instead of a human reading hundreds of pages, you upload a contract and our pipeline takes over:

  • PDF Validation: Validates the file to ensure it's a real legal document.
  • Document Extraction: Uses Nutrient DWS to extract the core structured data (values, deadlines, company info).
  • OSINT Due Diligence: Triggers a real-time web scrape via SerpApi to check if the contractor has a recent history of fraud or stalled projects in the news.
  • AI Copilot Analysis: Feeds all this context into an LLM via OpenRouter to get a predictive risk verdict.
  • Legal Document Generation: Uses Doctavian to map that verdict into a .docx template, instantly generating a standardized, print-ready PDF report for the government.
  • Public Transparency: Provisions on-demand public transparency domains using the live Name.com Core API.

How we built it

The frontend is a React SPA built with Vite, TypeScript, and Tailwind CSS to keep the dashboard snappy and clean. Our backend runs on Java 17 and Spring Boot 3, handling all the heavy lifting, file validations (via Apache PDFBox), and API orchestrations. Everything is tied to a PostgreSQL database hosted on Supabase.

We leaned heavily into the sponsor tech to make the pipeline work: Nutrient DWS for document extraction, SerpApi for the OSINT background checks, OpenRouter to route our prompts to Llama 3, and Doctavian to handle the final document generation. We also secured our transparency portal domain using Name.com.

Challenges we ran into

Getting five different external APIs to talk to each other in a single, synchronous 1-click workflow was definitely the hardest part. We had a tough time mastering the Doctavian document generation pipeline specifically - we had to carefully engineer the Spring Boot backend to make sure our JSON data payloads perfectly matched the template URNs in storage, otherwise the generation would time out.

Another big challenge was security. We realized late in the game that our AI API keys were exposed in the Vite environment, so we had to quickly refactor our architecture to isolate those calls on the server side to keep the frontend completely secure.

Accomplishments that we're proud of

Getting the entire end-to-end pipeline to work smoothly. Seeing a raw PDF go in, get parsed, cross-referenced with live Google News, judged by an AI, and printed as a beautifully formatted legal report in just a few seconds feels like magic.

What we learned

We learned a ton about Document AI extraction and template-based document pipelines. Juggling multiple third-party APIs while bridging a React frontend with a secure Spring Boot backend was a huge crash course in cloud-native system architecture and debugging under pressure.

What's next for Civis RJ

The MVP is working, but we want to take it further. The next step is integrating this directly into municipal transparency portals so citizens can actually see these AI-generated risk reports in real-time, completely opening up the black box of government contracts.

Built With

Share this project:

Updates

Submission history