Reopen

What it does

Many civic complaints get marked resolved while the problem is still there. I read reports from Pune and Bengaluru where citizens said their tickets were closed without action. Reopen checks the closure.

You give it your complaint and the closing note from the authority. You add photos of the spot. A Nemotron vision model compares them and gives a verdict with a confidence score. It also lists why the photos could mislead.

Then Tavily searches for your city's own grievance portal and appeal rules. A Nemotron reasoning model turns that into a dated plan. The plan has three steps. Reopen the complaint. Appeal to a higher officer. File an RTI request. Every step has a draft message you can copy and a source link. You can download the dates for your calendar.

What inspired me

I like building tools for problems around me. Jowar and Medbridge came from that. Reopen comes from the same place. A closed ticket should mean a fixed problem.

How I built it

The server uses Python and FastAPI. The page uses plain HTML and CSS and JavaScript. The server calls Nebius Token Factory for both models and Tavily for search. Nothing is stored on the server. Photos are checked by file type and size. The browser reads the photo time and warns you when it is old or missing. There are Privacy and Terms pages and a button that deletes the saved case.

Challenges

Getting access to the platform took effort because the billing step asked for a card. Finding the right model IDs took a few tries. Windows setup caused errors that I had to fix one by one. The hardest part was prompting. I wanted the model to say unclear instead of guessing. I also wanted every deadline to come from a real source. If no source gives a deadline the plan says so.

What I learned

Reliable JSON from a model needs strict prompts and careful parsing. Sources must be checked against what the search returned. A tool that admits its limits is easier to trust.

Limits

Reopen does not file anything. You review and send every message yourself. The photo verdict is an estimate and not proof. The demo link points to the public repo. You can run it with the README. The video shows a real run.

What is next

Neighbours could add evidence for the same spot. That would give a ward level report on closures. Serverless Jobs could send deadline reminders. A hosted version would let anyone try it.

Built With

• css
• fastapi
• html
• javascript
• python