Inspiration
Every year, millions of tenants face harassment, illegal eviction threats, and unfair lease terms. They often fold because they don't know the law, and legal counsel costs hundreds of dollars an hour. We realized that while "Legal AI" exists, it's usually just a text chatbot. Real confrontations happen in audio (face-to-face arguments) and vision (physical paper notices). We built CivicSight to be a multimodal "Pocket General Counsel" that gives power back to the tenant when they need it most.
What it does
CivicSight is a comprehensive legal defense system with three core "senses":
- Ears (Sonar Threat Detector): The app listens to real-time verbal confrontations. Using Gemini's multimodal audio capabilities, it detects illegal coercion (e.g., self-help evictions) and provides immediate verbal "Power Phrases" for the user to de-escalate the situation.
- Eyes (Omni-Vision Scanner): Users can scan physical lease agreements or eviction notices. The app identifies predatory "trap clauses" (like illegal entry rights) that violate specific local housing codes.
- Voice (The Legal Drafter): Unlike passive apps, CivicSight takes action. It uses Gemini to draft formal, legally binding PDF notices (Cease & Desist, Demand for Repairs) based on the user's specific GPS location and jurisdiction, ready to be emailed instantly.
How we built it
We built CivicSight using React Native (Expo) for a cross-platform mobile experience. The brain of the operation is the Google Gemini 3 API.
- Multimodal Integration: We pipe raw audio data and base64 image strings directly into Gemini 2.5 Flash for low-latency analysis.
- Location-Aware Prompting: We use
expo-locationto inject the user's city and state into the system prompt. This forces Gemini to act as a local expert, citing specific state statutes (e.g., "Texas Property Code") rather than generic advice. - Fail-Safe Engineering: We implemented a custom "race condition" handler for GPS data to ensure the app defaults to "General Legal Mode" if a satellite signal is weak, preventing crashes during emergencies.
- Document Generation: We convert Gemini's text output into formatted HTML and use
expo-printto render professional-grade PDFs on the device.
Challenges we ran into
The biggest challenge was handling latency vs. accuracy. Real-time audio analysis is heavy. We had to optimize the audio recording settings to balance file size with clarity for the Gemini model. Additionally, "hallucinations" in legal advice are dangerous. We solved this by using strict system instructions that force the model to cite the specific law code it is referencing, or admit if it doesn't know, rather than guessing.
What's next for CivicSight
We plan to implement a "One-Click Lawyer Handoff." While AI is powerful, a human review is the gold standard. In the future, users will be able to send their "Evidence Locker" (recordings, scans, and generated PDFs) directly to partner legal aid organizations with a single tap, creating a highly qualified lead for pro-bono lawyers.
Log in or sign up for Devpost to join the conversation.