Kiren Srinivasan’s independent practice

One stuck piece of work, finished.

Fixed-scope research, reliable agent workflows, and interface systems for founders who need a concrete result this month. You get the artifact, the checks, and the manual. No retainer required.

Wild output.
Manual discipline.
Draft first.
Verify before release.

Start small

A paid test before a broad build.

$1,500 to $2,500

72 hours after inputs are complete. One bounded question answered with evidence, risks, a first useful slice, and a fixed follow-on scope.

Inspect the Field Test

Three public offers

Scope you can point at.

Each offer names the finished artifact, starting price, delivery target, exclusions, and the checks that define done.

Compare every scope

Human-native and agent-readable

No second set of rules for machines.

People get a normal interface. Approved agents get documented access when it helps. Both follow the same scope, permissions, price, and review steps.

The manual

Readable by a person.

  • Clear seller, scope, price, and exclusions
  • Keyboard and no-JavaScript paths
  • Visible approval, cancellation, and support
  • Files, checks, limits, and handoff notes

The manifest

Callable without guessing.

  • Stable identifiers and structured records
  • Typed inputs, outputs, and errors
  • Read-only before write
  • Same terms as the visible human page

Selected public work

Proof lives in the source.

No invented client outcomes. Each story shows what exists, how it can be checked, and what was not measured.

See every proof record

Public repository, versioned release

webular

An open-source web research CLI that routes focused capabilities through a deterministic task graph.

Visible proof: Public source and release history.

Inspect the record

Public repository, validation workflow present

Visual Design System Extractor

A portable Agent Skills package that converts visual evidence into a validated design-system contract.

Visible proof: Portable skill package structure.

Inspect the record

Public repository, validators documented

Hermes Toolbox

Public, reusable agent building blocks with explicit sanitization, identity-neutrality, and safety checks.

Visible proof: Dry-run installation path.

Inspect the record

The working method

Gate. Build. Verify. Hand off.

The process is written so the buyer can see where scope, access, judgment, and release change hands.

Read the full method
  1. 01

    Name the finish line

    You should know the finish line before I touch the work. We name the file, screen, decision, or working path that must exist at handoff.

  2. 02

    Inspect the field conditions

    I inspect the actual sources and permissions before asking for more access. That is where constraints and failure cases usually surface.

  3. 03

    Build the smallest deep solution

    I build the smallest version that answers the question or makes the path work. Protocols and polish wait until they earn their place.

  4. 04

    Try to break it

    Then I try the obvious breakpoints: bad input, a rejection, an interruption, and a repeat request. The agreed checks decide whether it is ready.

  5. 05

    Hand over the artifact and the manual

    You get the artifact, its source, the checks I ran, the limits I found, and a manual someone else can follow.

One accountable operator

AI can do the first pass.

AI can help me reach a strong first pass faster. I still decide what is in scope, what evidence counts, where access stops, and whether the work is ready to release. The files and manual must stand on their own after the model session is gone.

AI helps with production. Kiren makes the final call on scope, evidence, exceptions, and release.