TaniJaga

Inspiration

In Indonesia, 77.5% of MSMEs operate without formal financial statements. Smallholder farmers focus heavily on production and marketing, leaving cost accounting neglected due to limited accounting knowledge, perceived complexity, and high costs. During harvest price crashes—such as the January 2026 chilli price crash in Karanganyar where farmgate prices fell to Rp 9,500/kg—farmers often sell below their actual Break-Even Price (BEP) of Rp 9,907/kg simply because they do not know their cost floor.

Furthermore, 62% of Indonesian farmers lack a verifiable transaction history, locking 17.3 million smallholders out of formal credit (like Bank BRI/Mandiri KUR loans) and trapping them in pre-harvest middleman debt. We built TaniJaga to turn forgotten paper receipts into instant cash, a defensible price floor, and bank-ready credit profiles.

What it does

TaniJaga creates an incentivised cost ledger loop for smallholder farmers:

Snap & Earn: Farmers photograph receipts from seeds, fertiliser, fuel, and labour expenses to receive instant cash rewards.
Verify & Classify: AI-based image processing scores photo quality and authenticity (1–10) to flag duplicates or altered images, automatically sorting expenses into COGS, OPEX, and CAPEX.
Know Your Floor: Aggregates expenses against yield in real time to present a live break-even price per kilogram.
Spoken Ledger: Uses ElevenLabs to deliver spoken plain-language ledger confirmations in Indonesian/English.
Bank-Ready Ledgers & Ground-Truth Insights: Automatically generates audit-ready financial statements for bank loan officers while building a macro-level supply data feed for enterprise buyers and lenders.

How we built it

We engineered a minimal, production-ready stack designed for technical evaluation rigour:

Frontend: Next.js 16 + Tailwind CSS & Lucide Icons deployed on Vercel.
Backend: Python FastAPI + Pydantic deployed on Railway.
AI Engine: Gemini 3.6 Flash for multimodal receipt quality scoring and expense classification.
Voice Synthesis: ElevenLabs Multilingual v2 for spoken ledger output.
Database & Storage: Supabase (PostgreSQL 15) synced live to the expense ledger.

Challenges we ran into

Receipt Quality & Fraud Prevention: Real-world agriculture receipts are often handwritten, crumpled, or poorly lit. Gemini 3.6 Flash accurately scores photo legibility and flags screen photos or altered images without rejecting valid receipts required by structured prompt boundaries.
Localised Voice Output: Integrating ElevenLabs Multilingual v2 to pronounce local agricultural terms, numerical costs, and currency values naturally in Indonesian.
Simplifying Accounting: Translating complex financial concepts (COGS, OPEX, CAPEX, BEP) into a frictionless zero-extra-tap workflow for farmers.

Accomplishments that we're proud of

Production-Ready & Fully Deployed: Delivered live HTTPS endpoints with zero localhost setup and clean Docker containerization for one-click reproduction.
End-to-End Multimodal Flow: Successfully linked vision-based expense classification with localised voice feedback and real-time BEP calculations.

What we learned

Incentives Drive Data Logging: Farmers do not need another complex accounting tool; they need an instant incentive (cash per upload) and a number they can trust before negotiating with middlemen.
Micro-Data Unlocks Macro Value: Individual expense receipts aggregated across smallholders create powerful ground-truth supply and risk intelligence for financial institutions and enterprise buyers.

What's next for TaniJaga

Gapoktan Cooperative Pilots: Deploying the incentivised upload flow with 1–2 Gapoktan cooperatives in Karanganyar, Central Java with real farmers and real payouts.
Institutional Partnerships: Packaging the auditable ledger as a credit-scoring data layer for KUR loan officers at Bank BRI and Bank Mandiri.
Commodity Expansion: Expanding beyond chilli to other high-value, high-volatility horticulture crops across Indonesia.