Market-Mama Haggler

Market-Mama Haggler a real-time Gemini-powered voice negotiator that interrupts, counsels, and counters like a seasoned market trader to secure better prices and instant savings for every shopper.

Comment

Inspiration: Market-Mama Haggler emerged from a recognition that traditional marketplaces operate as dynamic negotiation systems—emotion, tone, and strategy embedded into economic exchange. Modern e-commerce removed that interactive layer, replacing performance with static pricing. The project re-introduces strategic bargaining through AI, transforming price discovery into an engaging, culturally expressive, and economically empowering interaction engine.

What It Does: Market-Mama Haggler is a real-time multimodal negotiation agent that: Interprets product pages through visual analysis (extracting price, product data, coupon fields). Processes live voice input, detecting both offer value and tonal intent. Generates adaptive counter-offers in culturally styled Nigerian Pidgin. Highlights actionable UI elements (where to paste codes, click buttons). Tracks and displays cumulative savings in real time. Auto-produces a short, shareable victory clip after successful negotiation. It is not a chatbot. It is a personality-driven negotiation engine that merges commerce, performance, and automation.

How We Built It: Market-Mama is architected as a cloud-native, serverless microservice ecosystem on Google Cloud. Frontend: React/Next.js with WebRTC for low-latency streaming, Tailwind CSS for UI structure, and Framer Motion for animated interactions. Backend (Cloud Run): FastAPI-based services communicating via WebSockets and internal APIs.

Core Microservices: Orchestrator (session control & routing), Vision Analyzer (product data extraction with Gemini Vision), Price Intelligence (comparison and savings computation), UI Action Engine (DOM mapping & highlight positioning), Live Agent (Gemini Live API integration with dynamic contextual prompts), Clip Renderer (async media assembly via FFmpeg + Cloud Tasks)

Data Layer: Firestore (session & savings logs), Redis (active session + vision caching), Cloud Storage (temporary media assets with auto-deletion),

Security & Deployment: Cloud Run auto-scaling, Cloud Build CI/CD, Secret Manager for API keys, Least-privilege IAM roles.

Challenges We Ran Into: Maintaining sub-800ms round-trip latency for real-time conversation. Synchronizing interruption handling between frontend and backend. Achieving high accuracy across diverse e-commerce layouts. Engineering culturally authentic Pidgin tone generation. Designing clean data flow across six independent services. Ensuring demo-grade stability with fallback mechanisms.

Accomplishments We’re Proud Of: Fully multimodal bargaining agent (Voice + Vision + Action). Sub-800ms real-time interaction latency. High cultural authenticity validated by native speakers. Demonstrated measurable savings (₦45,000 in under one minute). Scalable, cost-efficient serverless architecture. Automated viral clip generator for social sharing. Polished, animated UI delivering immersive user experience.

What We Learned: Real-time conversational AI requires interruption support for immersion. Personality design significantly increases user engagement. Combining vision and voice enables contextual commerce automation. Latency optimization is critical to preserving conversational flow. Cultural authenticity demands collaborative iteration. Serverless infrastructure provides scalability without idle overhead.

What’s Next for Market-Mama Haggler: Multilingual and multi-persona expansion (regional dialect variations). Direct marketplace integrations for one-click negotiated checkout. Browser extension for universal shopping support. Savings analytics dashboard with social comparison features. Group negotiation sessions for bulk purchasing leverage. Affiliate revenue integration for sustainable monetization. Market-Mama Haggler represents the evolution of commerce AI—from static pricing assistance to personality-driven negotiation intelligence.

Built With

  • cloud-storage-sdk
  • cloud-tasks-client
  • fastapi
  • ffmpeg
  • firebase-authentication
  • firebasesdk
  • firestore
  • firestore-sdk
  • framer-motion
  • framermotion
  • gemini-live-api
  • gemini-multimodal-live
  • gemini-vision-api
  • google-cloud
  • google-genai-sdk
  • next.js
  • orchestrator
  • python3.11
  • react
  • recordrtc
  • redis
  • tailwind
  • tailwind-css
  • typescript
  • webrtc
  • websockets
  • zustand
Share this project:

Updates