# AgriBridge *A trilingual, AI-powered platform connecting farmers to schemes, subsidies, loans, and markets.* ### Project Brief AgriBridge is a trilingual, farmer-facing Android application that I conceived, designed, and built end-to-end while working as an AI Specialist & Data Scientist at 1M1B Foundation, developed for a partner organization to serve farming communities. I owned the complete product lifecycle single-handedly — UI/UX design, native Android development, backend architecture, AI integration, and cloud deployment — managing the entire build, from initial requirements through to production support, on my own. At its core, AgriBridge functions as a digital companion for farmers, bringing together six essential services in one place: an AI-powered conversational assistant for farming-related questions, a state-and-district-level government scheme finder, agricultural loan information, a marketplace for selling produce, live mandi price tracking, and government subsidy details — all delivered in the farmer's language of choice among English, Hindi, and Kannada. ### Problem Statement Farmers navigating government support and market information are typically faced with a fragmented, inconsistent landscape: schemes, loan facilities, and subsidies are spread across numerous state and district-level sources, many of which are difficult to search, irregularly updated, and rarely available in a farmer's native language. As a result, eligible farmers often miss out on schemes and subsidies they qualify for, struggle to find fair mandi prices before selling their produce, or depend on intermediaries who don't always act in their interest. This is made worse by the fact that most existing agricultural digital tools are English-first, leaving a large population of regional-language speakers underserved. What was needed was a single, reliable application that could pull live, verified information from disparate government and market sources and present it simply, in the farmer's own language, without requiring them to search across dozens of separate websites. ### Solution Architecture AgriBridge was structured as a three-layer system: a native Android client, a Python FastAPI backend, and an AI-driven data layer powered by the Gemini API. The Android application was built natively using XML-based layouts within a single-activity architecture, organized around the MVVM (Model-View-ViewModel) pattern to keep the interface cleanly separated from business logic — a structure that made the app easier to extend as new modules were added over time. On the server side, a FastAPI backend handled request routing and business logic, backed by MongoDB for persistent storage of scheme, loan, subsidy, and market data. Rather than relying on static datasets, I used the Gemini API as an intelligent scraping and crawling engine, continuously pulling live data from official state and district government sources and agricultural market listings — so the information farmers saw stayed current without manual upkeep. To keep the experience fast across three languages, I introduced a caching layer combining local, on-device caching with Redis: once a response was fetched in one language at runtime, switching languages afterward served the same data instantly from cache rather than triggering a fresh API call, removing redundant load and effectively eliminating wait time on language switch. The entire system was deployed and managed on a single AWS EC2 instance, configured with DNS support and a self-managed MongoDB instance, giving me full control over the production environment from provisioning through ongoing operation. ### Technical Execution I built and integrated six core modules: an AI chatbot for open-ended farming questions, a state-and-district-specific government scheme finder, a government agricultural loan directory, a marketplace module for listing and selling vegetables and fruit, a live mandi price tracker, and a subsidy information module reflecting each state's current offerings. Because the underlying data was sourced live through Gemini-driven web crawling rather than static datasets, accuracy verification was a central part of the build. I set up a cross-verification process that checked scraped results against multiple state sources and against Gemini's own outputs, which consistently held at close to 98% accuracy; the small remaining margin traced back to occasional source mismatches across states, which I could identify and account for rather than treat as pipeline errors. Multilingual support across English, Hindi, and Kannada was built around a caching-first design: the first API response for a given query was cached both locally and in Redis, so that subsequent language switches served pre-fetched content instead of re-querying the backend — removing loading delays and cutting unnecessary API usage. On the infrastructure side, I provisioned and configured the AWS EC2 instance, set up DNS routing, and self-managed the MongoDB deployment behind the backend — carrying the project from architecture decisions through to production deployment and ongoing support, without a dedicated backend or DevOps team. ### Results & Impact AgriBridge moved from concept to a fully deployed, production-supported application, built and delivered entirely by one engineer across design, Android development, backend architecture, AI integration, and cloud infrastructure. The live-data approach, verified at close to 98% accuracy through cross-checking across states, meant farmers received scheme, loan, subsidy, and mandi-price information that reflected real, current conditions rather than static or outdated listings — directly addressing the fragmentation that made this information hard to access in the first place. The caching-based multilingual architecture made the app equally usable for English, Hindi, and Kannada speakers with no performance penalty for switching languages, widening its accessibility to farmers who would otherwise have been excluded by an English-only interface. By consolidating six previously scattered services into a single application, AgriBridge reduced the number of separate sources a farmer needed to consult, lowering the effort required to access entitlements, loan information, and fair market prices. ### Key Learnings Owning a product end-to-end, from architecture decisions through to production support, sharpened my ability to make pragmatic trade-offs quickly — including recognizing early where caching would matter most, once repeated language-switch API calls stood out as an obvious inefficiency. Working with AI-driven live data extraction reinforced the importance of building verification into the pipeline rather than trusting model output on its own; cross-checking Gemini's scraped results against multiple state sources is what made a 98% accuracy figure something I could stand behind rather than assume. Designing for multiple languages from the outset, instead of retrofitting translation later, showed me how much an early architectural decision — like a caching layer — can determine whether a feature like multilingual support feels seamless or sluggish to the end user. Finally, managing the full lifecycle solo, from a partner organization's requirements through to a live AWS deployment, strengthened my end-to-end ownership mindset: architecting not just for functionality, but for something that could be maintained and operated independently once the initial build was complete.