Skip to main content
Initializing AgriRisk Engine
AgriRisk UZ v1.0 • Stage: Idea / Early Prototype 0%
Uzbekistan • Agricultural AI & Finance • Current Stage: Idea / Early Prototype

Data-Driven
Agricultural Finance

AgriRisk UZ turns every field in Uzbekistan into a digital “Field Passport” with AI-based Yield & Risk Scores. Banks get a clean, explainable risk view; farmers get access to collateral-free loans. In the next stage, these portfolios become regulated, tokenized assets for institutional investors.

0 M+
Hectares of arable land in Uzbekistan
AI
Yield & credit risk modeling engine
Scroll to explore the problem & solution

01 · Problem → Solution

Uzbekistan's fields are productive.
The financial signal is missing.

Agriculture is ~20% of Uzbekistan's GDP, but millions of small and dehkan farmers cannot access formal credit because land is state-owned and data on field performance is not digitized. We turn each field's agronomic history into a transparent, bank-ready risk profile.

The Collateral Trap

In Uzbekistan, farmland is owned by the state and most small and dehkan farmers do not have bankable collateral. Even when they manage their fields well, banks cannot "see" their yield history or risk profile. As a result, productive farmers are rejected or underfinanced, and agriculture remains dependent on informal credit and state support.

Loan Application #492 Rejected

Reason: No land collateral. Field performance not measurable in current systems.

The Digital Field Passport

AgriRisk UZ replaces static collateral with a living, AI-generated profile for every financed field. We combine satellite imagery, weather, and agronomic data into Yield & Risk Scores that banks can plug directly into their credit processes. Over time, each farmer builds a transparent performance history that unlocks better, cheaper finance.

  • AI-powered yield prediction & trend stability per field
  • Field-level Risk & Volatility Scores for bank risk models
  • Longitudinal “Field Passports” with yield, compliance & interventions
Expected impact

More farmers become “visible” and bankable inside the formal credit system.

+20–40%

Modeled increase in bankable farmers* (to be validated in pilots)

02 · Implementation & AI Engine

We translate satellite pixels into
bank-grade risk metrics.

Our pipeline ingests multi-spectral satellite imagery (Sentinel-2, Landsat), combines it with weather and soil data, and runs it through time-series and computer vision models. The output is a Field Passport: Yield Score, Risk Score, and Volatility Index that can be plugged directly into bank credit workflows.

1

Data Ingestion & Field Mapping

We map farmer fields via polygons and match them with historical satellite imagery, weather time series, and (where available) yield records. This data is stored in a geospatial database (PostGIS) as the foundation for AI scoring.

2

Time-Series & Vision AI Modeling

Time-series models (LSTM, Temporal Fusion Transformer) learn the relationship between NDVI/NDRE curves, weather patterns, and historical yields. Computer vision models (YOLOv8, SegFormer) detect stress and disease zones from imagery. Together they generate Yield, Risk, and Volatility scores per field.

3

Financial Scoring & Bank Integration

We convert agronomic signals into bank-native metrics such as Probability of Default (PD) and Loss Given Default (LGD) bands. A web dashboard and REST APIs let banks use these scores during loan origination, monitoring, and reporting. In later phases, they also power structured and tokenized “Harvest Notes.”

Implementation Stack & AI Tools

Backend & Data
Core
  • Python, FastAPI / Django
  • PostgreSQL + PostGIS (geospatial)
  • Data pipelines with Celery / Redis
AI & Analytics
Models
  • PyTorch, TensorFlow
  • LSTM, Temporal Fusion Transformer
  • YOLOv8, SegFormer for imagery
Frontend & UX
Dashboard
  • React / Next.js
  • Tailwind CSS
  • Mapbox / Leaflet for maps
Infrastructure & Tokens
Later phases
  • Docker, CI/CD (GitHub Actions)
  • Cloud hosting (AWS/GCP/local)
  • Permissioned EVM chain & regulated tokenization (licensed CASPs)

This is the target stack for the prototype & pilot. Exact tools will adapt to bank requirements, local hosting constraints, and regulator feedback.

Demo farmer avatar
FIELD PASSPORT #UZ-8821
SIMULATED

DEMO DATA • FERGANA VALLEY

Risk Tier
A
Score: 92/100
Satellite view of agricultural fields
LAT: 40.3821 LON: 71.7823
NDVI Index
0.78 +12%
Proj. Yield T/HA
12.4
Risk Model
LOW A-TIER

03 · Workflow & Usage

From Field
to Financial Asset.

This is how AgriRisk UZ will work in practice once deployed: digitize fields, score them with AI, and turn them into structured, investable pools for banks and institutional investors.

Scroll horizontally on desktop
01

Digitize the Field

Farmers or field agents map land parcels through a simple web or mobile interface. Our system retrieves historical satellite imagery, weather, and vegetation indices for each polygon and creates an initial Field Passport.

02

Score with AI

Time-series (LSTM/TFT) and computer vision models (YOLOv8/SegFormer) analyze vegetation patterns, climate risks, and crop health. The system outputs Yield, Risk, Volatility, and Maturity scores that are visible in the bank dashboard and accessible via API.

03

Finance & Structure

Partner banks use Field Passports to approve loans and monitor risk in real time. In later phases, portfolios of these loans are bundled into “Harvest Notes” and, under Uzbekistan’s regulated digital asset framework, tokenized for institutional investors and development funds.

04 · Roadmap & Current Stage

Execution Timeline

Current Stage: Idea / Early Prototype
Stage: Idea / Early Prototype

01

0–6 Months

Prototype & Baseline Models

Goal: Build a functional demonstration of the AgriRisk UZ concept and validate that banks can understand and use field-level AI scores.

Key actions:
• Implement the first version of the bank dashboard UI (this demo).
• Integrate open Sentinel-2 satellite and basic weather datasets for a small number of demo fields in Uzbekistan.
• Train baseline LSTM models for NDVI-based yield forecasting using public datasets and synthetic yield labels.
• Design the data model for a “Field Passport” (Yield Score, Risk Score, Volatility Index, Maturity Index).
• Conduct interviews with at least one risk officer and one agronomist to refine UX and metrics.

IN PROGRESS

Stage 2 · Pilot with Bank Partner (Prototype → MVP)

Goal: Run a real-world pilot with a partner bank and a portfolio of farmers to prove that AI-based Field Passports are accurate, usable, and compliant with local processes.

Scope & participants:
• One partner bank (or MFO) + 50–100 real farmers across 2–3 regions and 2–3 crop types.
• Mix of small and medium farms so models see different management styles and risk patterns.

Data & AI work:
• Connect anonymized historical loan and repayment data (where available) to field polygons.
• Collect yield, crop type, and basic management data through simple forms / field agents.
• Retrain and calibrate time-series and CV models using local data to improve yield and risk predictions.
• Implement periodic model retraining pipeline and performance monitoring (e.g. R², MAE for yield, AUC for default/no-default).

Bank integration & UX:
• Embed AgriRisk scores in the bank’s workflow – either via a lightweight API or downloadable reports used by credit officers.
• Co-design a minimal “credit memo” format where field scores map to PD / LGD bands the bank already understands.
• Add simple portfolio views (by region, crop, risk tier) in the dashboard for risk managers.

Governance & compliance:
• Define data-sharing agreements, anonymization rules, and retention periods with the partner bank.
• Document model assumptions and limitations in language understandable to compliance teams.

Success metrics (examples):
• Prediction error for yield below a defined threshold (e.g. MAE < 20%).
• At least a 10–20% increase in credit access for participating farmers vs a control group, with no deterioration in NPLs.
• Positive qualitative feedback from credit officers (scores are understandable and helpful in decisions).

Stage: Prototype / Pilot (Upcoming)
Target: 6–18 Months

02

Stage: Prototype

Target: 18–36+ Months

03

Stage: MVP → Launched

Stage 3 · Tokenized Harvest Notes & National Scale

Goal: Turn AgriRisk UZ into core risk infrastructure for agricultural lending, and gradually open agriculture as a regulated, tokenized asset class for institutional capital.

Product & tech evolution:
• Evolve the pilot into a full MVP used in production by at least one bank on hundreds of loans per season.
• Implement role-based access control, audit logs, and robust API documentation for multiple bank integrations.
• Add alerting and monitoring for fields and portfolios (e.g. drought alerts, disease spikes, yield downgrades).

Regulation & compliance:
• Engage with local regulators (e.g. CBU, NAPP) and licensed crypto-asset providers to participate in a digital asset sandbox.
• Design “Harvest Notes” as diversified pools of farmer loans where underlying risk is transparently scored by the AI engine.
• Ensure all tokenization is done via licensed intermediaries, fully KYC/AML-compliant and aligned with Uzbekistan’s legal framework.

Capital & ecosystem:
• Offer banks and DFIs (development finance institutions) a new way to co-finance agriculture with clear risk/return data.
• Allow investors to choose portfolios by region, crop type, and risk band, while farmers continue to receive classic loans in soum.
• Explore integration with insurance partners to build risk-pooling and index-based insurance products based on the same field data.

Scale & impact metrics:
• Multiple banks onboarded, thousands of fields monitored per season across several regions.
• Documented reduction in manual underwriting time for banks (e.g. 20–30% less time per application) and better portfolio visibility.
• Demonstrable improvement in farmers’ access to finance and more stable, predictable cash flows at the sector level.

Long-term Vision: MVP & Launched Phases

05 · Team

Two engineers building agri-finance infrastructure.

Founders registered in AI500!
Full-stack + AI-native + Domain Focused.

A

Aliakbarxon Nurillaxonov

Co-Founder • AI & Backend

Software engineer focused on backend systems, geospatial data, and AI. Leads the AgriRisk AI engine, orchestrating time-series forecasting and ML services deployment.

Python FastAPI PyTorch PostGIS

Why we are the Right Team

AI & Data Infrastructure

We don't just use APIs; we build pipelines. From geospatial data engineering to training time-series models, we own the full data lifecycle.

Full-Stack Delivery

We design and ship real web applications. We bridge the gap between complex ML outputs and intuitive dashboards for non-technical bankers.

Regulatory Realism

We are excited by LLMs, but disciplined by reality. We design systems that fit Uzbekistan’s banking regulations and data availability.

Compact & Agile

Two founders who code. We iterate rapidly based on user feedback, bypassing bureaucratic delays to reach product-market fit faster.

Immediate Focus

Building a credible prototype for expert critique. Next step: Onboarding risk officers and regulatory advisors to evolve AgriRisk UZ beyond the hackathon.

Join the Pilot

Ready to explore
AI-powered agri finance?

Get in Touch

For the AI500! Stage 1, this is a demo website with simulated data. The next step is to connect real fields and start a small pilot with a partner bank.