GetHired.devBuild notes for getting hired
Search/
Menu

Niyam

A policy-to-code verification project designed around one boundary: software can collect evidence and propose a result, but a human should approve consequential actions.

The engineering problem

Policy documents are written for people, while enforcement systems need explicit, testable conditions. A naive converter can make a confident mistake and hide the reasoning. Niyam separates source text, extracted requirements, generated checks, evidence, and approval so a reviewer can inspect each step.

System map

Policy evidence with a human decision boundary

Niyam keeps source, interpretation, executable checks, and approval separate so a generated answer never becomes policy merely because a model produced it.

  1. 01Source policyPreserve the exact passage and document identity.
  2. 02Requirement extractionPropose testable requirements with visible uncertainty.
  3. 03Constrained checkGenerate and run an inspectable verification step.
  4. 04Human approvalA named reviewer accepts or rejects consequential output.

Project evidence snapshot

Niyam is a hackathon project, not an autonomous compliance product. There is no responsible runtime benchmark to publish yet, so this panel reports review-boundary facts instead.

5
Separated stagesSource, requirement, check, evidence, approval.
1
Mandatory human gateConsequential changes require approval.
Top 30%
Hackathon placementContext, not proof of correctness.
0
Autonomous approvalsA deliberate product boundary.

The product boundary

  • Preserve the source passage behind each interpreted requirement.
  • Generate a proposed check without treating generated code as approved policy.
  • Run verification in a constrained environment and retain the result.
  • Show uncertainty and missing evidence instead of inventing certainty.
  • Require an explicit person to approve a consequential change.

What the project demonstrates

The useful portfolio signal is not “AI reads policy.” It is the review boundary: provenance, deterministic checks, failure visibility, and a human decision record. The project reached the top 30% of its hackathon field, but that placement is context rather than proof of technical correctness.

Honest limits

Niyam is a project and demonstration, not legal advice or an autonomous compliance system. Generated interpretations require domain review, and a passing check does not prove that an organization satisfies every obligation in a real policy.

Code and author

See Aarav Chandel's GitHub profile for currently public project work and implementation evidence.