Intelligence • Structure • Scale

Ballast

Stability beneath intelligence.

An independent AI product studio in India, run by one builder. The model handles language. Everything a user has to trust (pricing, limits, permissions, safety) is deterministic code underneath it.

Stability beneath intelligence SYSTEM VIEW
Stability beneath intelligence Language, intent and context flow along paths into a single processing node. That node crosses a dashed boundary into a deterministic block holding rules, state, limits and audit, which rests on a solid base.

Intelligence

Language on the surface. The model does the part that genuinely needs judgement about words, and nothing beyond it.

Structure

Rules you can read, test, and audit. If a behaviour matters, it is specified in code, not coaxed out of a prompt and hoped for.

Scale

Built to hold. Multi-tenant, observable, and predictable about how it degrades when a model or a dependency fails.

01

In build

Nothing shipped yet
Coming soon

The first product

In build. It applies the studio's split, language on the surface and deterministic rules underneath, to a domain where being confidently wrong is expensive. It gets a name, a subdomain, and a page here when there is something real to use.

Want to know when it ships? Put me on the list →
02

How I build

The studio's point of view
01

Clarity over noise

Say exactly what a system does and what it refuses to do. A product that cannot tell you which half of itself is allowed to be wrong has not been designed, only assembled.

02

Systems thinking

AI drafts; a person approves. Approval gates, consent, and revocation are first-class in the schema, not bolted on after a demo. Every product is a system before it is a screen.

03

Built to last

Durability over novelty. Products go in front of real users early, with clear success and kill criteria, and the market sets the next move, not my own optimism.

04

Practical intelligence

Intelligence has to earn its place. If a deterministic rule does the job, it gets the job; the model is for the part that genuinely needs language, not for the parts that are merely easier to prompt than to specify.

03

Who builds it

One builder

Rishabh Shukla

AI Systems Architect · Founder of Ballast

AI Lead by day, architecting production AI systems for enterprise teams: multi-tenant agentic platforms, permission-aware retrieval, natural-language analytics over enterprise data, and the observability and guardrail infrastructure that keeps them safe to operate at scale.

Ballast is where that work becomes whole products rather than one layer of someone else's platform, under a single point of view: the AI should be delightful to talk to and boring to depend on.

04

Work with me

What the studio takes on

What I take on

  • Architecture review of an AI system already carrying real traffic: where determinism belongs, how tenants stay isolated, how it degrades.
  • Building a production AI product end to end: retrieval, guardrails, evals, observability.
  • Turning a working prototype into something multi-tenant, observable and operable.

What I don't

  • Staff augmentation or hourly body-shopping.
  • A chatbot over your documents with no deterministic layer under it.
  • Anything that needs 24/7 on-call ownership from one person.