NearLens

• 

💡 Inspiration

In many communities, people have smartphones but lack access to reliable, context-aware information about their surroundings — from identifying local products to discovering nearby services.

We wanted to create a tool that turns vision into insight, bridging the gap between what people see and what they can know. NearLens was inspired by the idea that “seeing should lead to understanding.” Whether identifying an item, finding where to buy it nearby, or translating local signage, NearLens makes the visual world more intelligent and locally relevant.

⚙️ How We Built It

NearLens is designed as a multi-agent workflow with Gemini-powered reasoning to ensure intelligent decision-making at every step:

Intro Agent – Welcomes the user, interprets the initial query, and sets task context.

Vision Analyzer Agent – Processes images to detect objects, text, and scenes.

Local Recommender Agent – Leverages geolocation and contextual insights to fetch nearby recommendations using google places API.

Translator Agent (optional) – Converts output into the user’s preferred language for inclusivity.

All agents communicate via a central Sequential Orchestrator, which uses Gemini models for reasoning, prioritization, and error handling across multiple agent outputs.

We deployed both:

Backend (FastAPI, Python) → on Cloud Run

Frontend (Next.js, React) → on Cloud Run

This serverless setup enables instant scalability with minimal ops overhead. Agents communicate through ADK messaging channels and Cloud Run endpoints.

☁️ Tech Stack

Google Cloud Run (frontend + backend)

Google ADK (Agent Development Kit)

Gemini models for agent reasoning

FastAPI (Python) for backend

Next.js (React) for frontend

Google Places API for location-based insights

Cloud Build for CI/CD automation

🧠 What We Learned

Designing a multi-agent architecture that works in sequence with context-aware reasoning.

How Gemini models can enhance decision-making and prioritization in AI agents.

Deploying frontend and backend independently on Cloud Run while ensuring seamless integration.

Integrating AI outputs into a real-time, location-aware user experience.

🚧 Challenges

Ensuring smooth communication between multiple agents without message conflicts.

Handling real-time image analysis and geolocation recommendations within latency constraints.

Deploying both frontend and backend on Cloud Run with consistent routing and environment configurations.

Training agents to reason effectively with Gemini across diverse contexts.

🚀 What’s Next

Add offline caching for low-connectivity regions.

Expand Gemini reasoning capabilities for deeper contextual insights.

Extend NearLens into tourism, accessibility, and local commerce applications.

Explore voice-assisted and AR-based visual guidance to enhance user interaction.

Built With

• cloud-build
• cloud-run
• fastapi
• gemini
• google-adk
• google-places
• nextjs