3L

• 

  ask

• 

  find location

• 

  generate business ideas

About the Project 🚀 Inspiration Starting a business is hard—especially when you don’t know the local market. We wanted to make it radically easier for anyone, anywhere, to discover the best business to open in their neighborhood, using the power of AI and real-time data.

🧠 What We Learned LLMs are powerful, but need careful prompt engineering to extract structured, actionable insights from unstructured text. APIs for geocoding and web search can be unreliable or rate-limited; fallback strategies and error handling are essential. User experience matters: Real-time feedback and clear, actionable results are key for trust and engagement.

🛠️ How We Built It Frontend: Built with Expo React Native for cross-platform mobile support. Clean, modular UI with animated status and map feedback. WebSocket pipeline for real-time analysis updates.

Backend: FastAPI microservices in Docker Compose. Orchestrates a multi-stage pipeline: Extract location from user query (LLM) Geocode to coordinates (OpenStreetMap/Apify) Find nearby businesses (Google Maps/Apify) Analyze neighborhood profile (LLM + web search) Generate business ideas (LLM, structured output)

AI/LLM: Used OpenAI GPT and Gemini via LangChain for extraction, research, and ideation. Structured output enforced with Pydantic schemas and OpenAI function-calling. Prompt engineering to ensure reliable, context-aware results.

Data: Combined LLM reasoning with live data from geocoding, Google Maps, and Brave Search APIs.

🧩 Challenges Dependency hell: LangChain’s rapid evolution and conflicting dependencies (e.g., langchain-apify vs. langchain-google-genai) required custom Docker build steps and careful version pinning. LLM output validation: LLMs sometimes output creative or non-standard JSON; strict schema validation and fallback parsing were needed. API rate limits and failures: Apify and Brave Search quotas were hit during testing; we implemented error handling and fallback strategies. Real-time UX: Ensuring the mobile app always shows a clear, single status line (not a log), and that users get instant feedback even for long-running AI tasks.

🌍 Impact With 3L, anyone can get instant, data-driven business recommendations for any location in the world—no MBA required.

Built With

• apify
• brave
• docker
• docker-compose
• expo.io
• fastapi
• langchain
• python
• react-native
• websocket