CivicLens

Logo of the project
Layout of the webpage

Inspiration

Democracy dies in bureaucracy. In most local governments, public meetings last for hours, and the "ground truth"—official budgets and legal PDFs—often exceeds 500 pages. We realized there is a massive accountability gap: it is humanly impossible for a citizen to cross-reference a politician’s spoken promise in real-time against a complex financial document. We built CivicLens to turn Gemini 3 into an autonomous, multimodal auditor that ensures what is said at the podium matches what is written in the books.

What it does

CivicLens is an agentic transparency dashboard. A user uploads a video of a city council meeting and an official budget PDF. Using Gemini 3’s native multimodal understanding and "High" reasoning levels, the app:

Watches & Takes notes: Identifies factual claims regarding spending, dates, and policy from recoded videos of meetings.
Deep Reasoning: Scans the provided documentation to verify the claim.
Generate Analysis: Flags discrepancies (e.g., "The Mayor promised $1M for schools, but the budget only shows $100k") with timestamps and direct quotes from the source PDF.

How we built it

This project was built using the "Vibe Coding" philosophy. Instead of manual boilerplate, we used Google AI Studio to steer Gemini 3 Pro into generating the application architecture.

Multimodal Engine: We leveraged Gemini 3’s ability to process video and audio natively without separate transcription layers.
High Reasoning (Thinking Level): We utilized the thinking_level: high parameter to ensure the model could handle the complex logic required to compare abstract spoken promises with granular financial data.
Frontend: A clean, React-based dashboard designed for high-stakes data visualization.

Challenges we ran into

The biggest challenge was "Contextual Nuance." Sometimes a speaker uses slang or refers to a previous meeting. We solved this by providing Gemini 3 with a "memory" of the document and instructing it to use its reasoning tokens to confirm whether a discrepancy was a lie or just a different line item.

What we learned

We learned that the barrier between "Idea" and "Software" has vanished. By using Gemini 3's Vibe Coding capabilities, we were able to focus 90% of our energy on the impact and logic of the auditor and only 10% on the syntax.