Inspiration
We understand how important privacy is when dealing with sensitive information. We've seen how exposing the wrong information can put people at real risk, and journalists, legal-aid teams, and human-rights researchers face similar risks when protecting themselves and their sources. We want to bring that experience into building an LLM that puts privacy first.
What it does
Zireh is a privacy-first collaborative space where investigative teams can create and work on cases while having access to the research speed and intelligence of AI models without having their work compromised on external servers. Your work is stored on your computer. Zireh lets local AI prepare sensitive questions for cloud AI, which is powered by Nvidia Nemotron, while encryption and unlinkable payments powered by Zcash reduce what each outside service can link to you.
How we built it
Zireh is a shared workspace for teams investigating sensitive cases. Before a question goes to AI, Nvidia Nemotron on the user's computer tries to hide names and other identifying details; case files are encrypted before they sync with teammates. We also added privacy routing and a way to pay without linking each AI request to a person.
Challenges we ran into
The hard part was protecting people's identities without making questions so vague that AI becomes unhelpful. AI can miss names or clues, so we added extra checks that can rewrite or block a risky question, but no system can promise perfect protection.
Accomplishments that we're proud of
Teams can collect evidence, share observations, connect clues on a case board, review findings together, and prepare a conclusion. We're proud that privacy is part of the research flow, with a log that shows what information was sent to AI.
What we learned
Protecting privacy takes more than removing names: places and other details can also identify someone. We also learned that people should be able to see what the AI received and check its work.
What's next for Zireh
To keep improving how easily teams can share a case and reach a conclusion together.
Log in or sign up for Devpost to join the conversation.