Gurshabd

Building @useDefinite (YC S26) Prev @BitGo CS @uwaterloo

[Gurshabd @GurshabdV]

https://t.co/FR1ZEjjrlx

[Gurshabd @GurshabdV]

DeepSeek V3 just beat GPT-5 at financial reporting…

And all we did was add a compiler. Code doesn’t ship with error because the compiler catches them first so we did the same for financial reports. Our compiler checks every number against the real records, and sends the wrong ones back in a feedback loop.

More accurate, ~50x cheaper

[Gurshabd @GurshabdV]

We're Definite. Our agents run the entire regulatory reporting pipeline for banks, from raw ledger data to the filed report.

[Gurshabd @GurshabdV]

super excited to launch @UseDefinite - the most accurate AI agents for financial reporting

Definite automates regulatory reporting for banks end to end, from raw ledger data to filed report, turning weeks of manual work into hours.

https://t.co/jnaZ7iM47P

[Gurshabd @GurshabdV]

@itsmoni_02 Whenever the usage limits reset

[Gurshabd @GurshabdV]

@GohilHardy Definitely this guy iykyk

[Gurshabd @GurshabdV]

@RajaPatnaik @TryLedgix Finally: Enterprise reality is multi-cloud. Companies don't want three different compliance stacks for Bedrock, Vertex, and Foundry.

@TryLedgix is the unified Compliance Gateway that stays consistent across the stack.

[Gurshabd @GurshabdV]

@RajaPatnaik @TryLedgix Most cloud logs are a post-hoc narrative

Ledgix is a deterministic interceptor. We capture the reasoning loop before the action is signed

If a regulator asks "Why?", you show them a tamper-evident execution artifact that the cloud provider physically cannot retroactively edit

[Gurshabd @GurshabdV]

@RajaPatnaik @TryLedgix In 2026, "marking your own homework" is a Control Failure.

Under new compliance frameworks, companies need Independent Validation. If your audit trail is managed under the same root credentials as your AI agents, you’ve lost your organizational independence.

[Gurshabd @GurshabdV]

@RajaPatnaik @TryLedgix The way we like to put it is they provide the court, the jury, and the stenographer. Ledgix is the independent witness.

They offer Observability: logging what the AI claims it did.

We provide Explainability + Verifiability: cryptographic proof of the deterministic logic path.

[Gurshabd @GurshabdV]

@ssbrouhard @badlogicgames Zero trust is the only realistic framework. We're building @TryLedgix to be that deterministic layer. It provides mathematical proof and signed artifacts of agent actions. Verification shouldn't just be a harness; it should be a prerequisite for production.

[Gurshabd @GurshabdV]

@_BananaCrystal The 11-day delay is the trust gap. Finance won't give agents signing authority or a wallet without verifiable compliance proof. Building @TryLedgix to be that gateway: providing mathematical proof of agent actions so we can move money at the speed of the agent.

[Gurshabd @GurshabdV]

@vasuman Completely agree: auditability isn't a feature, it’s a prereq. I’ve building @TryLedgix to solve this: a compliance gateway that provides evidence that actually satisfies a legal audit.

If you can’t prove why and how an agent took an action, it shouldn't be in production.

[Gurshabd @GurshabdV]

@emmajo They’re just execution shy

[Gurshabd @GurshabdV]

@justbyte_ All I know is ThinkPads in elementary school used to hit

[Gurshabd @GurshabdV]

@AICEOGiuliano @JustJake Super cool - How are you all handling the flakiness in your evaluation suites? I find that getting a stable baseline for regression testing is so much harder with agents. Are you leaning more into golden datasets or LLM-as-a-judge for that validation?

[Gurshabd @GurshabdV]

@PremChauraisya Scratch but even that might be kinda iffy rn

[Gurshabd @GurshabdV]

@RussellQuantum The challenge is that even with all that, the audit step is still manual. I've been working on @TryLedgix trying to solve that, so that prod-ready doesn't just mean it doesn't crash, but it’s compliant enough for a legal audit.