Fuell AI / Fuell LLM

A language model for the way Indian businesses work.

Fuell LLM is our own large language model, now in development. It is designed to understand bookings, orders, bills, rent and stock, and to work across English, Hindi and more Indian languages.

One platform. Every part of your business.Connected by design.

Design goals

Smaller, sharper,
built for real work.

General chat models know a little about everything. Fuell LLM is aimed at the tasks businesses run every day, at a cost that makes sense for them.

Fluent in operations

Understand orders, bookings, invoices, rent ledgers and stock movements, and the way teams talk about them.

Indian languages

Designed for English, Hindi and the Hinglish people actually type, with more languages planned.

Efficient to run

Sized to answer quickly and affordably, so AI features can reach small businesses.

Grounded by design

Built to work with RAG and tools: answer from supplied sources and cite them.

Careful with actions

Trained to ask for approval before changing records or contacting customers.

Private deployment

Planned options for running the model where your data and rules require.

Coming soon · In developmentFuell LLM is not available yet. Nothing on this page is a release date or a performance claim.

Roadmap

Where we are, and what comes next.

We will share progress at each phase. Research partners see results first and help decide what we build next.

  1. Phase 1Now

    Research & data

    • Business-domain tasks and evaluation sets
    • Licensed and consented training data
    • Model size and serving cost targets
  2. Phase 2

    Train & tune

    • Base model training or adaptation
    • Instruction tuning for operations tasks
    • English, Hindi and Hinglish coverage
  3. Phase 3

    Partner preview

    • Limited API access for research partners
    • Scored against our evaluation tasks
    • Safety review before wider access
  4. Phase 4

    Inside Fuell

    • Powering Don AI, automation and RAG
    • Developer API
    • Private deployment options

How we will measure it

Judged on business tasks, not trivia.

We are building evaluation sets from real operational work, and plan to share results on them, including where the model falls short.

  • ExtractionPull the right fields from orders, bills and booking messages.
  • Grounded answersAnswer from supplied sources, cite them and decline when they do not cover the question.
  • Multilingual understandingThe same task asked in English, Hindi and Hinglish.
  • Tool useChoose the right action, fill it correctly and ask for approval.
  • SafetyRefuse harmful requests and protect personal information.
  • Speed and costResponse time and cost per task on realistic hardware.

Research partners

Help shape Fuell LLM.

Who can join?
Restaurants, hotels, PGs, retailers and developers who want AI that fits their work, and are happy to share feedback.
What do partners get?
Early access to previews, a direct line to the research team and a say in which tasks we prioritise.
Do I have to share data?
No. Sharing example data is optional, and only happens under a written agreement about how it is used.
Will Fuell LLM be part of the Fuell platform?
That is the goal. Once it meets our quality bar on business tasks, we plan to offer it inside Fuell alongside other models.

Build the next model with us.

Tell us about your business and the AI tasks that matter most to you. We will be in touch as partner places open.