Samrit Mukherjee

🚀 The Problem It Solves Creating professional-looking banners, posts, and advertisements is often a challenge for: 💼 Startups that can’t afford professional designers 🧑‍🎓 Students and creators with limited design experience 👨‍💻 Individuals who lack the creativity or time to design from scratch 🎨 Non-designers overwhelmed by complex tools like Photoshop While AI image generators (like ChatGPT or DALL·E) can generate great results, they output static images that are not editable. If users want to change a small part (like text or layout), they have to regenerate the entire image — losing consistency and control. 💡 Cosmic Canvas Fixes This Cosmic Canvas gives users the power to: ✨ Generate images using AI prompts 🧩 Get smart layout and caption suggestions 🖌️ Edit every element individually — like in Photoshop or Canva 🧠 Customize AI-generated designs without starting from scratch It bridges the gap between AI creativity and hands-on customization, making stunning design accessible, editable, and intuitive for everyone.

KrishakSarthi — AI-Powered Agriculture Assistant for Every Farmer Tagline From smartphone to WhatsApp — one AI companion covering crop planning, disease detection, and market access for India's 140 million farming households. The Problem Most Indian farmers don't have access to timely agronomic advice. Agricultural extension officers are stretched thin, internet literacy is low, and existing agri-apps require a smartphone and stable connectivity. Farmers lose crops to diseases they could have prevented and miss market opportunities because they can't find buyers. What We Built KrishakSarthi is a full-stack AI agriculture platform with three deeply integrated layers: Mobile App (React Native + Expo) A farmer-friendly Android/iOS app with large touch targets, high-contrast green UI, and full trilingual support (Hindi, Bengali, English). AI Crop Disease Detection — photograph a diseased leaf; Gemini Vision identifies the pathogen, severity, and treatment within seconds Crop Prediction Engine — combines live weather data with soil and location inputs to recommend the best crop for the season Crop Planner + Calendar — step-by-step guided farming plan with a visual activity calendar Document Analyzer — farmers photograph government forms, seed packets, or invoices and get plain-language summaries AI Chatbot — RAG-powered conversational assistant backed by a curated agriculture knowledge base Farmer-Buyer Marketplace — supply chain screen connecting farmers directly with verified buyers; live GPS location sharing for last-mile logistics Daily Routine Tracker — task scheduler tailored to crop growth stages RAG Knowledge Backend (Node.js + Vercel) A Retrieval-Augmented Generation API that turns agriculture PDFs and documents into an intelligent Q&A engine. PDF ingestion pipeline chunks and embeds agricultural documents into Pinecone vector store Queries are semantically searched, top chunks retrieved, and answers synthesized by Gemini with source citations Exposes a clean REST API (POST /api/chat) consumed by both the mobile app and WhatsApp service Multi-language translation layer for Hindi ↔ English responses WhatsApp Bot (Node.js + Meta Cloud API) For farmers who only have a basic WhatsApp account. Hooks into Meta's WhatsApp Cloud API via webhook Farmers ask agriculture questions in Hindi or English and get back RAG-powered answers — no app download needed Natural-language Crop Planning: farmer types "5 acre mein basmati dhan, Sodepur, 15 June se" — the bot extracts all fields, confirms, generates a full 90-day farming plan PDF, and delivers it back over WhatsApp Voice call support for farmers who prefer speech Challenges We Solved Zero-connectivity farmers: WhatsApp channel needs only basic 2G; no app install required Multilingual NLP extraction: Free-form Hindi messages parsed by Gemini to extract structured crop fields (crop name, area, location, date) reliably PDF over WhatsApp: Generated PDFs uploaded to Meta Media API and delivered as WhatsApp documents, not just text Encoding pipeline: Built a robust ingestion pipeline to chunk, embed, and upsert heterogeneous agriculture PDFs with metadata preserved for source citations Impact Potential Targets 140M+ smallholder farmers in India who rely on WhatsApp daily Reduces crop loss from undetected disease with sub-10-second AI diagnosis Bridges the farmer-buyer gap directly in the app, cutting out middlemen Works in the language the farmer is most comfortable in Challenges we ran into RAG queries took 30+ seconds, exceeding Twilio's hard limit Solution: Implemented async job queue with polling mechanism—respond immediately to Twilio, process in background, caller polls for result Audio Processing Pipeline WhatsApp Integration Challenges Multi-Turn Conversation State Keeping conversation context across multiple message exchanges for crop planning Handling WhatsApp message retries (duplicate IDs) Solution: Redis-based session storage with auto-expiration Language Support (Hindi/English Mix) WhatsApp media upload constraints (file size limits) Proper multipart form-data formatting for PDF uploads Solution: PDF generation with size optimization and chunked delivery Rate Limiting & Quota Management Handling API rate limits from multiple providers simultaneously Managing quota exhaustion gracefully Solution: Multiple API key rotation and fallback models