Projek Candi Presents, SuaraTani

SuaraPintar
KomunitasTani
Home
BelajarTani
PasarTani

Inspiration

All of our members were born and raised in Indonesia. Witnessing how the country has grown, we can’t help to see that the contribution of farmers all over the archipelago has been immense, especially in the economic sector. Therefore, inspired by the challenges faced by Indonesian smallholder farmers, particularly their limited access to crucial agricultural information in this fast-paced industry and market connections due to literacy and digital literacy barriers that many farmers face. We recognized the potential of voice AI to bridge this gap. Our goal was to empower farmers with knowledge, improving their livelihoods, and creating a solution that is truly accessible, intuitive, and impactful for the backbone of Indonesia's economy

What it does

SuaraTani ("Farmer's Voice") is a voice-interactive mobile application designed to be a virtual agricultural advisor and market linkage tool for Indonesian smallholder farmers. Farmers can use simple voice commands (or text entry) in Bahasa Indonesia to ask questions about a wide range of agricultural topics, including common crop diseases, local weather forecasts, optimal farming practices for specific crops, and current market prices for their produce. The system then leverages AI to generate accurate, localized advice and information, which is delivered back to the farmer in clear, high-quality Bahasa Indonesia via ElevenLabs' Text-to-Speech. Furthermore, a basic module facilitates connections between farmers and local buyers or cooperatives, aiming to improve market access and fair pricing for their produce.

How we built it

We built SuaraTani with a mobile-first, low-bandwidth optimized user interface (UI), specifically addressing the internet access constraints in rural areas. The core of our system is a backend logic that efficiently processes farmer queries that are generated by Bolt AI and personally refined by our team. For voice input, we integrated ElevenLabs' Speech-to-Text (STT) to accurately transcribe Bahasa Indonesia voice commands. These transcribed queries are then fed into a general-purpose Large Language Model (LLM) for initial advice generation, which is further augmented by our local knowledge base. This knowledge base contains specific Indonesian agricultural information, such as common pests by region and planting calendars, all stored in a structured database. Crucially, the AI-generated advice is then converted back into clear, understandable Bahasa Indonesia speech using ElevenLabs' high-quality Text-to-Speech (TTS). Our front-end is designed for extreme intuitiveness, requiring minimal digital literacy. But due to limitations on resouce and time, we substitute the model with gemini API and Eleven Labs as the text-to-speech interpreter.

Challenges we ran into

The first problem we faced was adapting Bolt to build the basic structure of our app. Although it's one of the most powerful tools we've ever come across for building a working app in a short amount of time, it requires hefty and sometimes complex debugging that demands quite advanced expertise. Moreover, one of our primary challenges was ensuring the accuracy and contextual relevance of agricultural advice for diverse Indonesian regions and crop types. We considered many aspects that could make certain crops optimal for the market, such as price and climate compatibility, which presented quite a challenge in deciding the best prediction method for crop types. This required extensive research and the development of a robust local knowledge base to augment the general LLM. Another significant hurdle was optimizing the application for low-bandwidth conditions, necessitating a very lean and efficient UI. Integrating ElevenLabs' STT for Bahasa Indonesia also presented its own set of challenges in ensuring precise transcription across various accents and speech patterns. Finally, designing an interface that is truly intuitive for users with limited digital literacy requires rigorous user experience considerations and simplified interaction flows. Finally, finding a farmer who wants to use the app and give feedback is an interesting yet really hard experience for our team.

Accomplishments that we're proud of

We are incredibly proud to have developed a working prototype that is already being used by real farmers in Indonesia; this demonstrates the power of voice AI in addressing critical societal needs. Successfully integrating ElevenLabs' Speech-to-Text and Text-to-Speech capabilities in Bahasa Indonesia to create a seamless voice interaction experience is a major accomplishment. We are also proud of our efforts to build a localized knowledge base, which is crucial for providing actionable and trustworthy advice to farmers. Most importantly, we've created a solution with high potential for significant social impact, empowering farmers and strengthening rural economies. We hope that this could be a new path for combining and collaborating the knowledge gained by local farmers with the versatility and practical use of AI, together boosting the economy of Indonesia.

What we learned

We would like to thank StackBlitz and Bolt for making this competition possible, and giving us a chance to create something new like SuaraTani. This hackathon experience reinforced the importance of understanding the unique challenges of the farmers and tailoring solutions to their needs accordingly. We learned invaluable lessons about designing for accessibility and inclusivity, especially in contexts with limited digital literacy. The process also highlighted the critical role of robust data and localized knowledge in making AI solutions truly effective and trustworthy. Furthermore, we gained deeper insights into optimizing web applications for low-bandwidth environments using new innovative tools such as Bolt, ElevenLabs, and many more frameworks that we used.

What's next for Projek Candi

I think we can develop this app further with just a little more time, anyways we're just playing around with great minds. Our greatest challenges is our own schedules and agendas that if solved, we can make this app reach the true people