Karthik RameshContact
All workLive synthetic demonstration

ReplayOS

A workflow control plane that converts operational evidence into typed, approval-gated execution with recovery and verifiable traces.

For
Operations and product teams automating consequential, failure-prone workflows.
Ownership
Product strategy, Workflow model, Control architecture, Full-stack delivery, Evaluation design
System
FastAPI, React, TypeScript, PostgreSQL, Policy engine
ReplayOS control room showing an approval-gated operational workflow and evidence trace
ReplayOS control room with synthetic workflow evidence.Synthetic demonstration data
The workflow problem

What had to change

High-stakes automation becomes dangerous when consequential actions are hidden inside model-generated prose and failures cannot be reconstructed.

The product response

How the product behaves

ReplayOS compiles operational evidence into inspectable workflows that people can approve, execute, recover and replay without giving the model direct authority.

User journey

From first signal to inspectable outcome

  1. 01Collect evidence
  2. 02Propose a workflow
  3. 03Apply deterministic policy
  4. 04Approve material actions
  5. 05Execute idempotently
  6. 06Verify and replay
Architecture

The control path

The model participates inside a product boundary. Typed software, policy and human review retain authority where consequences matter.

  1. Evidence

    Structured events and operational context

  2. Compiler

    Typed workflow proposal and preconditions

  3. Control

    Policy, approvals and idempotency

  4. Execution

    Bounded effects and postcondition checks

  5. Replay

    Recovery, evidence export and hash-chained traces

Product decisions

Choices that shape the system

01

Models propose

Typed software retains authority over permissions and effects.

02

Recovery is first-class

Partial failure is expected and represented in the workflow.

03

Evidence survives execution

Operators can understand what happened and why.

Evaluation and reliability

What keeps the product inspectable

Controls

  • Approval gates
  • Idempotent effects
  • Postcondition verification
  • Deterministic evaluation path

Verified evidence

  • Live synthetic demonstration
  • Recovery workflow
  • Hash-chained execution traces

Current limitations

  • The public deployment is a synthetic operational demonstration, not a payment processor.
  • External effects are mocked and no production customer data is used.
Recorded synthetic demonstration of the ReplayOS workflow.Synthetic demonstration data
Karthik’s ownership

Product strategy · Workflow model · Control architecture · Full-stack delivery · Evaluation design

GitHub repository Open live demonstration View all work