Inspiration

We watched people pass by those in need every day, caught between compassion and caution. In our own city, we saw potholes remain unfixed for months despite countless complaints. The problem wasn't a lack of goodwill, but a crisis of trust. We were inspired to build a bridge—not another complaint box, but a system where every act of help could be verified, tracked, and turned into collective hope. We asked: what if technology could engineer trust?

What it does

Sahayak transforms passive observation into verified action. When a citizen encounters a person in need, they don't just give cash or walk away. They use Sahayak to record a 30-second video interview. Our AI, powered by Gemini 3, analyzes the video, understands the stated need ("I'm hungry," "My child needs medicine"), and creates a dignified, verifiable profile. This profile is instantly routed to nearby NGOs. When an NGO provides help (a meal, medicine), they upload proof. Our AI audits the match between the need and the provided aid, generating a public, tamper-proof audit trail. For civic issues like potholes, it assesses severity and verifies repairs, creating accountability.

How we built it

We built entirely within Google AI Studio using React. Instead of a traditional backend, we engineered a suite of specialized Gemini 3 agents:

The Field Verifier: Processes video/image to create structured profiles from street encounters.

The Trust Auditor: Cross-references needs with resolution proof, outputting match certificates.

The Civic Engineer: Analyzes infrastructure photos for severity and repair verification.

The Hope Synthesizer: Generates positive stories from successful actions for the Unity Feed. The React frontend orchestrates these agents, managing state and presenting the verified data through a clean, map-based dashboard.

Challenges we ran into

Our biggest hurdle was prompt engineering for real-world chaos. Getting Gemini to consistently extract actionable needs from noisy street videos—despite background traffic, varied accents, and emotional speech—required hundreds of iterations. We also initially struggled with managing context across multiple agent calls until we implemented Thought Signatures to maintain reasoning continuity. Finally, designing a UI that made AI's complex verification process simple and transparent for end-users was a significant design challenge.

Accomplishments that we're proud of

We're most proud of creating a functional trust engine. In testing, Sahayak successfully routed 15 simulated cases with 100% audit accuracy, proving that AI can be the impartial verifier in social systems. We built a complete application without a traditional backend, showcasing Gemini 3's power as a reasoning layer. Our prototype demonstrates that technology can address human problems of distrust at scale.

What we learned

We learned that multimodal AI is about understanding context, not just classifying objects. The difference between a "person on street" and a "person seeking a blanket" is in the audio and temporal understanding. We also learned that building for social good requires extreme technical precision—one wrong transcription could misrepresent someone's dignity. Most importantly, we learned that the "Action Era" means building systems where AI orchestrates real-world outcomes, not just generates text.

What's next for Sahayak

Next, we will integrate Gemini Live for real-time, voice-guided NGO worker surveys in the field. We plan to develop predictive need mapping by analyzing case location patterns to proactively deploy NGO resources. We are in talks with municipal corporations in Pune and Bengaluru to pilot the civic infrastructure module. Ultimately, we envision Sahayak as a public utility for verified compassion—a nation-building platform where every citizen can be a certified helper.

Built With

• backend
• gemini