The messy jobs.
In your stack.
Then off your plate.
Repeating work gets a real workflow. Built in your tools. Handed to your team.
Six years as a software engineer at
Kustard Technologies
Software Engineer2020–2022
Navi
Senior Software Engineer2022–2026
GoodScore
Senior Software Engineer2026
Solutions
If it happens every week, it should not need a hero.
Most internal AI stalls at a chatbot on a pile of PDFs. The work here starts from the job — then a system your team actually uses, in your tools, with a handoff they can own.
01
Ops and workflow agents
Ops, people teams, anyone drowning in intake.
Approvals, routing, follow-ups, and handoffs live in Slack, email, and someone’s memory.
Agents and workflows. Humans stay in the loop where judgment still matters.
Done meansWork moves from intake to done. Not a demo that chats.
02
Support and internal knowledge
Support, IT, and teams tired of answering the same forty questions.
Tribal knowledge. “Check Confluence.” A ping instead of a search.
Retrieval over your docs, tickets, and wikis — with sources and access control — in Slack, the helpdesk, or a small internal UI.
Done meansAnswers people trust enough to stop pinging each other.
03
Reporting without the scramble
Founders, ops leads, anyone who builds the same weekly pack by hand.
Metrics archaeology across databases, sheets, Stripe, the CRM.
Scheduled briefs and dashboards fed by real data, with the boring joins handled.
Done meansThe report arrives. Nobody assembled it this week.
04
Documents into the system of record
Finance, ops, anyone filing documents into a system you already pay for.
Invoices, emails, PDFs, and forms sitting outside the tools you already use.
Extraction, classification, and filing — with review for the messy cases.
Done meansThe document is in the right place, with the right fields. People only touch exceptions.
Fit
When this is the right hire
A fit when
- The same job happens every week
- The data and tools already exist
- Someone inside will own it after handover
- You want it running, not presented
Not a fit when
- You only want to “add ChatGPT”
- There is no workflow — only a vague interest in AI
- You need a research lab, not a system in production
- Nobody will take it after handover
Process
Name the job. Ship the system.
01
Work audit
Name the jobs that should not be manual — and which ones can actually run.
02
Design the system
Data, tools, human checkpoints, failure modes, and what “done” looks like.
03
Build and deploy
Product, backend, models or retrieval, infra — in your cloud, in your tools.
04
Handoff
Source, runbooks, evals. You keep the system. You don’t keep a contractor on the critical path.
What you leave with
A running system. Not a deck.
- A running system in your cloud or account
- Source you own
- Docs and runbooks
- Eval checks for quality, cost, and the ugly cases
Also
If the system needs a real product, that gets built too.
Frontend, backend, and infra — six years of that path. The system is the lead. The engineering is why it ships.
About
Still writes the code. Still owns the deploy.
Six years as a software engineer at product companies — shipping things people actually use. The work now is the repeating jobs inside companies, and the systems that run them.
Then off the critical path.

Contact
If the same job ate this week, send it.
India (IST). Replies usually land within a day.