Building Odisha's first AI layer, in public.
Kortiv AI is an independent project adapting existing AI models for Odisha's real problems — starting with cash crop disease detection.
No polish yet. Just the work. 🌾
01 / Flagship Project
Global plant-AI doesn't know Odisha's crops.
Most agricultural disease-detection models are trained on datasets from other continents — featuring different climates, soil chemistry, and disease vectors.
We are building directly on top of open-source vision foundations, but fine-tuning them with datasets native to Odisha. Our goal is to provide local farmers and agricultural extensions with diagnostic accuracy tailored specifically for our fields.
We are physically documenting, photographing, and classifying crop disease symptoms directly from local farms. No synthetic data, no shortcuts.
Data Pipeline Status
v0.1-rawAre you an agricultural researcher, botanist, or farmer in Odisha with leaf disease imagery?Contribute your dataset.
02 / Horizon
This is layer one. Here is where we are headed.
Faking completeness does not help. We are framing these as future roadmap items, ensuring transparency on what is live versus what is planned.
Cash Crop Disease Detection
Adapting computer vision systems for local crops like Cashew and Betel Vine to bypass external dataset bias.
Odia-first LLM (Shiva-Alpha)
Fine-tuning small, efficient language models trained specifically to understand administrative Odia, local dialects, and agrarian vocabulary.
Climate & Disaster Prediction Models
Adapting meteorological models to local micro-climates, focusing heavily on predicting cyclone impacts on coastal farm belts.
Governance & Policy Automation
Parsing regional state documents, local schemes, and land registry records to translate complex legalities into accessible plain speech.
03 / Context & Accountability
Built by someone from here, for here.
Kortiv AI started with a simple frustration: artificial intelligence tools built in California or Delhi rarely understand Odisha's micro-climates, soil chemistry, agrarian dialects, or local governance structures.
When global models attempt to detect crop diseases here, they frequently misclassify local variations because they lack regional training data. When language models process regional queries, they struggle with administrative Odia and local contexts.
We are building this openly because local problems require dedicated, specialized focus. Rather than waiting for external companies to adapt their generic models to our context, we are building the data layers and fine-tuned models locally from the ground up.
— Founder, Kortiv AI
Independent AI Layer Initiative
04 / Build Log
Field Notes & Development Log
No polished press releases. Just a chronological stream of the actual progress, data pipeline issues, model failures, and physical farming insight we gather along the way.
Moisture Distortion in Ganjam Cashew Farms
Collected 350+ raw cashew leaf samples. High ambient humidity in coastal Ganjam alters leaf sheen, which causes vision models to flag false glares. Adjusting training filter parameters to neutralize glare distortions.
Local Workstation over Cloud API
Provisioned basic localized fine-tuning hardware. Decided against cloud-based APIs to keep building costs sustainable. Local model runs are slower but allow us to maintain 100% control over dataset pipeline storage.
Baseline Accuracy Drop on Local Crops
Tested standard open-source vision classification baseline. Accuracy dropped by nearly 34% when encountering wild-type, unmanaged betel vines from Puri farms compared to pristine standard academic datasets. Local training is confirmed essential.