About

A small firm that would rather be right than busy

RebellionTech is deliberately small. One engineer, a limited number of concurrent engagements, and nobody standing between you and the person building your system.

Direct

No account managers

1:1

You talk to the builder

India

Working worldwide

Who builds this

One engineer, and no one to hide behind

You work directly with the person writing the code. There is no account manager relaying messages, no junior team the work quietly gets handed to, and no ambiguity about who is responsible when something breaks.

Aurin Desai

Founder & Chief Executive Officer

BITS Pilani — Computer Science, Advanced Machine Learning

Email directly

Engineer turned AI systems architect. I started RebellionTech because I kept being handed the same broken thing to fix.

A model wired to an API, a demo that impressed a boardroom, and a system that fell apart the moment it met a real workflow. The failure was almost never at the model layer. It was retrieval nobody tuned, evaluation nobody built, permissions bolted on at the end, and failure modes nobody had thought about until a customer found them.

So RebellionTech does the unglamorous part on purpose. I design the whole system — data through interface — measure whether it actually works, harden it against the cases that matter, and then hand you everything needed to run it without me. Being one person is a real constraint on how much I take on. It is also why nothing gets lost in translation.

Where the work goes

System architecture

Routing, state, tool execution and failure handling. The layer that decides whether a system survives its first real month.

Retrieval & knowledge engineering

Getting the right context in front of the model every time. Most hallucination problems are retrieval problems wearing a disguise.

Evaluation that survives reality

Benchmarks built from real traffic, scored per slice. Knowing whether a system is actually good is harder than building it.

Production reliability

Observability, drift detection, rehearsed rollback. Intelligence that cannot hold under load was never intelligence.

What this means for you. Fewer concurrent engagements than an agency would take, and a far shorter path between a question and an answer. If the work needs specialists I do not have, I will bring them in and tell you exactly who is doing what — rather than quietly subcontracting it.

How we think

Four positions we actually hold

Not values on a wall. These change what we do on a Tuesday.

Understand before you build

The expensive mistakes are made in the first two weeks, when nobody has written code yet and everybody is agreeing enthusiastically about a problem they have each understood differently.

The system is the product

The model is one component among many, and rarely the one that decides whether the thing works. Retrieval, evaluation and failure handling do more for outcomes than another few points of benchmark score.

Ownership is not negotiable

If a client wants to take the system and walk, they should be able to. Building in switching costs is a business model that only works on people who are not paying attention.

Say the uncomfortable thing early

That the data is not good enough. That the timeline is fantasy. That this does not need AI. Delivering that news late is the single most expensive thing a consultancy does.

Ethics

Where we draw lines

Specific commitments rather than a statement of principles, because specifics are the only kind you can be held to.

We check provenance before ingestion

Where data came from, what consent covers it, and whether you have the right to train on it. A model built on data you cannot defend is a liability, not an asset.

We score fairness per slice, not in aggregate

Wherever a system touches people — hiring, credit, access, care — aggregate accuracy hides exactly the failures that matter. Slice-level reporting is part of the deliverable.

We decline work we should not do

Surveillance of individuals without their knowledge, manipulation dressed as personalisation, and anything that implies clinical judgement without a clinician. This costs us revenue and the rule stands anyway.

We make refusal a designed behaviour

A system that says "I do not have enough to answer that" is more valuable than one that always produces something. Designing the refusal is as much work as designing the answer.

Start here

Find out what you actually need
before you spend a rupee building it.

Send us the problem in whatever form you have it — a paragraph, a deck, a half-working prototype. You get a written architecture audit back in 72 hours.

A written read on what you are actually trying to build

The architecture we would use, and the two we rejected

Where this fails in production, and what that costs

A realistic budget range and timeline — or a reason not to start

Request the auditaurindesai@gmail.com

No discovery call required · No obligation · You keep the document either way