AI in agriculture
Independent analysis of how AI is actually being used in Indian agriculture — what works, what doesn't, what's hype.
40 articles
-
Field memory: where agent memory meets the real world
Field Memory is meant for the messy reality of US agriculture: years of field history, scouting photos, recommendations, product evidence, and grower follow-through spread across people and software.
-
Two Desais on Krishi.System: where AI advice for farmers actually comes from
Pratik joined Sachi Desai and Venky Ramachandran on Krishi.System for a 30-minute deepdive on the gap between agronomy literature and how farmers actually speak. Two complementary perspectives on B2B agentic AI: US large-holding and Indian smallholder.
-
KissanAI is in India's 100 Top AI Start-ups, 2026
Indiaspora and Zinnov published their joint '100 Top AI Start-ups in India: The 2026 Report' in March. KissanAI is on the list. A short post on what the report tracks, why we are honoured to be in this particular cohort, and what the list says about Indian AI in 2026.
-
Today we are announcing FieldFoundry
FieldFoundry is the second product in the KissanAI portfolio. The AI Command Center for agriculture, built for North American commercial production. What it does, why we built it, and what 'context fragmentation' means when you try to fix it.
-
Five days at the India AI Impact Summit
We launched Dhenu Platform at Bharat Mandapam last week. The conversations that followed shifted how we think about the next 12 months. A team reflection on what we saw, who we met, and what the response from the IndiaAI community is telling us about agentic AI for agriculture.
-
IndiaAI Innovation Challenge: KissanAI receives the ₹25 lakh recognition
We've been recognised by the IndiaAI Innovation Challenge for our work on voice-first agricultural AI. A short note on what this funds, why the IndiaAI mission matters to us, and what we'll use the support for.
-
Dhenu Platform is live: agentic AI for agriculture, built on NVIDIA
KissanAI launched Dhenu Platform at the India AI Impact Summit 2026. The first agentic AI platform for agriculture, built on NVIDIA NeMo, shipping a 20-agent suite that lets any agribusiness deploy AI to farmers in days, not months.
-
How we shipped agentic harnesses for agriculture
An engineering essay on the agent harness work: multi-agent orchestration, tool use, validation loops, deep research. The path from one model answering one question to coordinated agents that plan, retrieve, validate, and act.
-
Writing on the Wall: How Overlooked Data Can Fortify Legacy Industries
A line of graffiti in Pompeii survived two thousand years of weather, war, and volcanic ash. Most enterprise data does not get that lucky. Agentic AI is the steward that finally takes those trivial signals seriously.
-
Trust in AI for agriculture starts with benchmarks
An essay on why public benchmarks, not glossy demos, are how AI advisory services earn the trust of farmers and policymakers. The premise behind AI AgriBench, what evaluation we are doing, and what we are deliberately not.
-
On the AIM 100 list, and what I think it actually signals
Analytics India Magazine has named me to its 2025 list of the 100 Most Influential People in AI in India. A note about who deserves the credit, what I think the list actually signals about Indian agriculture AI in 2025, and why I am quietly pleased.
-
22 languages is not 22 translations
KissanAI currently supports 22 languages across two clear tiers: voice in 12, including English, and text in 10 more. Here is what that coverage does and does not mean.
-
Reasoning SLM for agriculture: research preview
A research preview of two Qwen3 models tuned for climate-resilient agriculture in India: a 4B model and a 0.6B micro model trained on synthetic, agronomist-aligned dialogues.
-
A Maharashtra farmer's harvest with AI
An Economic Times feature profiled a Maharashtra farmer using KissanAI advisory for a full crop cycle. Numbers, not narrative. What changed in input cost, what changed in yield, and what the farmer told the reporter about how trust was built.
-
Building agriculture knowledge graphs from messy multilingual data
A methodology essay on the part of the work that nobody asks about: turning years of curated Indian agricultural content into a knowledge graph a model can actually reason over. Crops, pests, products, climates, languages, and the joins between them.
-
Pratik on the Nebius Spotlight at GTC 2025
Pratik joined the Nebius / GeekWire Studios spotlight at NVIDIA GTC 2025. A short conversation on the cloud infrastructure behind Dhenu: why Nebius, what the cost curve looks like, and what serving inference to Indian farmers demands of a GPU stack.
-
Founding the AI AgriBench consortium
Why we joined the Center for Digital Agriculture and the Extension Foundation in launching AI AgriBench, a benchmarking consortium for generative AI in agriculture. What it is trying to do, what we are contributing, and why agricultural AI needs this work.
-
The visibility gap nobody talks about
Agribusinesses cannot see what is happening at the farmer level. Farmers cannot see what agribusinesses know. The wedge that closes that distance is operational intelligence, the part of agricultural AI nobody is shouting about.
-
Why agriculture AI gets stuck at pilot
Most ag-AI pilots in 2023 and 2024 did not graduate to production. The reasons are structural (domain depth, the last mile, enterprise procurement) and better models are not fixing them. What the path past pilot actually requires.
-
What is actually new in Dhenu 2
The Dhenu2 family shipped last month. This is about the part I care most about, climate awareness: what changed in the data, the architecture, and the eval, and what it gets us on the ground.
-
What the Bay Area AgTech ecosystem looks like in late 2024
A field journal from the AgTech Alchemy meetup we hosted in October. Who showed up, what they are working on, where the conversation is going, and why a small Bay Area community of agritech founders and funders is starting to matter.
-
Why we joined the AI Alliance
We've joined the AI Alliance alongside AI4Bharat, IIT Jodhpur, Infosys, People+AI, Sarvam AI, and Wadhwani AI. A short note on why the open-AI thesis matters specifically for agriculture and what we are bringing into the alliance.
-
Voice-first, not chat-first
Every farmer-facing build at KissanAI ships with voice on by default, in every language we can serve well. Why access, not features, drives that call, and what it means for an agribusiness scoping a deployment.
-
KissanGPT got a shoutout on The Ranveer Show
KissanGPT got mentioned on Ranveer Allahbadia's TRS Hindi podcast during a conversation with Masters' Union professor Dr. Nandini Seth, and a clip of the moment is making the rounds. Nice when the product turns up in conversations we didn't seed.
-
Made-in-India LLMs and where Dhenu fits
The current crop of indigenous Indian language models, sorted into what is working, what is vapor, and where domain-specific models like Dhenu fit. Written to clarify positioning rather than to win an argument.
-
KissanAI joins NVIDIA Inception
We have been accepted to the NVIDIA Inception program. What this gets us, what it does not, and why we think the partnership matters for Indian agriculture AI.
-
Pratik on Emergent Behavior: growth strategy for an AI in farming
Pratik went on EB-10 with Cogniscendo. A candid conversation on the unglamorous parts of building KissanGPT: the silence when farmers do not know what to ask, mellowing growth before monetisation, and why farmer-side revenue does not exist in developing economies.
-
Three of us on the JioGenNext Founder Talk
Lokesh, Pratik, and I sat down together on the JioGenNext Founder Talk channel: the rare conversation that put all three KissanAI cofounders in the same recording, with the engineering, business, and founder voices answering different parts of the same question.
-
Open-sourcing Dhenu Llama 3
We are releasing Dhenu Llama 3, an Indic agricultural language model built on Llama 3 8B. What changed from Dhenu 1.0, what the model is now usable for, and why we are putting the next iteration of our work in the open.
-
Pratik on Microsoft #ScaleUpThursday
Pratik joined Microsoft Reactor's #ScaleUpThursday for the season's closing episode: the KissanGPT origin story, why voice beats apps for farmers, and what it actually costs to train and serve a 7B agriculture model.
-
Dhenu Vision: a multimodal model for the field
Dhenu Vision is an early multimodal model for identifying crops and visible field symptoms from photographs. This engineering note explains the curation, reasoning, and failure modes behind it.
-
AIISC: bridging knowledge gaps in agriculture with generative AI
Pratik gave this week's Founder Talk at the AI Institute of South Carolina: how a domain model for agriculture gets built from the corpus up, and what it takes to put it in front of farmers who would not otherwise reach extension.
-
Shortlisted in the IndiaAI Innovation Challenge
A short note that KissanAI has been shortlisted in the IndiaAI Innovation Challenge under the agriculture track. What the shortlist means, what it doesn't, and why the programme's focus on agriculture-specific applications is good for the ecosystem.
-
KissanAI and UNDP: a CoPilot for farmers
KissanAI and UNDP are partnering to develop a voice-based, vernacular CoPilot for climate-resilient agriculture. Here is what the partnership set out to build and who it aims to serve.
-
GPAI India Summit 2023: AI for the Global South
A trip-report essay from the GPAI Summit hosted in New Delhi this December. What was discussed about AI for the Global South, where domain AI for agriculture fit into the conversation, and a few personal observations from the floor.
-
Dhenu 1.0: training a domain LLM for Indian agriculture
What we announced this week at the GPAI Summit in Delhi, why the weights are going up on Hugging Face, and what we learned training a 7B-parameter model on three hundred thousand instructions written in the language farmers actually use.
-
Why generic LLMs miss agriculture
Five months into running KissanGPT in 2023, the limits of general-purpose language models in agriculture had a clear shape. These field observations explain why local data and domain systems mattered.
-
What six weeks of farmer questions taught us
An early field note from KissanGPT: the language and question patterns we saw, what farmers expected from the product, and what those conversations taught us to build next.
-
KissanGPT: a small experiment, in Gujarati
A career in Silicon Valley, an evening in Surat, and a bet that the farmer my father trained should be able to ask a phone questions in his own language. This is how KissanGPT started, what we built first, and what we learned in the six weeks after the launch.
-
Introducing KissanGPT: the first voice-based vernacular AI for Indian farmers
Today we are releasing KissanGPT, a voice-first, multilingual generative AI copilot for Indian agriculture. First of its kind in two ways: voice as the primary interface, and Indic-language support from day one.