Skip to content
Manish Kumar

Work

Things I build, and how I build them.

This is where the engineering shows: what my experience covers, the products I've built, and the decisions behind them.

Where something is still coming, it says so.

Practice

What my experience covers.

  • Cloud and reliability

    Leading the teams that build and run cloud platforms, where the job is keeping important systems correct and running.

  • AI-assisted engineering

    Putting AI agents to work in real codebases, as on this site, and building products around them, as with CreatorBrio.

  • Products and businesses

    Taking products from a rough idea to something people use, and working out how each one earns its keep.

Security

Security is part of how I build.

  • Secure delivery

    SBOMs, hardened container images, vulnerability scanning, security checks in CI/CD, and controls on infrastructure as code.

  • Cloud and runtime security

    Secure cloud configuration, secrets management, access controls, workload security and operational visibility.

  • Governance and evidence

    Security standards, audit evidence, clear security processes and accountable operational practices.

Projects

Products I've built.

FlashListing

Status: LiveVisit FlashListing(external site)
FlashListing's public sale page for the Oak Park Moving Sale, listing items with prices and availability.

The problem

Selling a lot of things at once, for a move, an estate or a garage sale, usually means a photo album, a spreadsheet only you can read, or a separate marketplace post for every item.

Why I built it

“Every time I moved, it went the same way.”

Read the full story on the FlashListing site (external site)

What works today

  • A public catalog behind one link, with several photos per item and availability that updates in real time.
  • An inbox for buyer inquiries, with no buyer account needed to browse or ask.
  • A free Basic plan and a paid Premium plan with more catalogs, featured items, follower alerts and collaborators.

Coming, not here yet

  • An AI-assisted listing workflow, a Business plan for estate liquidators and repeat resellers, and annual billing.

Decisions that shaped it

  • Buyers never need an account, so a shared link works for anyone who opens it.
  • It stays a catalog: no payments, no pickup arrangements and no transaction fees.

Fidem

Status: LiveVisit Fidem(external site)
Two example Fidem findings on an applicant's pay stub, one passed and one recommended for review, each with its evidence.

The problem

Independent landlords and small property managers usually check rental applications alone, by eye, which makes an inconsistent pay stub or bank statement easy to miss.

Why I built it

“Almost every independent landlord verifies applications the same way: alone, by eye.”

Read the full story on the Fidem site (external site)

What works today

  • Checks pay stubs, bank statements and, optionally, an employment letter and government ID for income, pay periods, employer, bank activity and ID expiry.
  • Shows the document values behind every finding.
  • Secure applicant uploads from an emailed request, with no account, and a review workflow with notes and an activity log.

Decisions that shaped it

  • It shows what the documents say and leaves the decision to the landlord. It doesn't contact employers or banks, or pull credit or background data.
  • Documents are encrypted in storage and in transit and never used to train AI, and it says plainly that it hasn't completed a SOC 2 audit yet.

CreatorBrio

Status: Open sourceVisit CreatorBrio(external site)
The CreatorBrio editor with a chat panel beside a carousel slide titled Three habits for a slower morning.

The problem

Designing on-brand Instagram carousels by describing what you want, with the brand applied to every slide, on your own machine.

Where it came from

It began as a fork of Open Carrusel, an MIT-licensed project, and is now maintained independently with the original copyright notice preserved.

What works today

  • A Claude design agent that creates, rewrites or removes slides from a chat panel, after a planning step works out the request's intent, audience and message.
  • Slides written as real HTML and CSS, rendered by the same code for preview and export, so the PNG matches what you saw.
  • Brand settings for colours, fonts, logo and style, several brands per installation, and Instagram's three carousel formats.

Coming, not here yet

  • Workflows for planning a series of posts, reusable branded templates, and platforms beyond Instagram.

Decisions that shaped it

  • Code, not the model, carries out the binding parts of a plan, so every failure is a typed outcome and nothing is silently substituted.
  • It's local-first: plain files, no accounts, no database, and chat goes to Anthropic through the user's own Claude Code account.

Hachira

Status: In development

AI governance built on evidence, not assumptions.

Hachira is a local-first AI governance control plane designed to help organizations understand their AI footprint, track governance decisions, and maintain traceable evidence for oversight and compliance workflows.

The emphasis is inventory, visibility, accountability and auditability. It supports governance and compliance work rather than replacing it, and it isn't public yet.

How I build

Built properly. Shipped early.

  • Design before code

    I'd rather spend a day on the data model and the boundaries than a week untangling the wrong shape later.

  • Design for failure

    Edge cases are part of the task. Deciding what happens when a call fails or the input is malformed is part of building the feature.

  • Ship early. Improve deliberately.

    Release the smallest version that delivers real value. Get the essentials right, including security, reliability, and critical workflows, then improve through real feedback instead of polishing indefinitely.

  • AI with accountability

    Agents do a lot of the typing. Tests, review and a person who owns the result decide what ships.

Working on something similar?

If you're working on a hard engineering or product problem, I'd like to hear about it.

Get in touch