About
A consulting practice built to ship, not to sell slides.
TechAegisAI is a founder-led practice putting agentic AI into mid-market operations. Real workflows, messy data, and a payback you can put on the next board update.
The founder
Suraj Malthumkar
Three years at AWS building the ML inside Bedrock and GuardDuty. Now I build agents that run in production, not slideware. Every engagement is scoped to a number, shipped in weeks, and handed to your team to own.
Founder's note
Hi, I'm Suraj.
I spent three years at AWS, on the machine learning inside two products you've probably used without noticing. GuardDuty, which sifts billions of security events a day looking for the one that matters. Then Bedrock, the platform enterprises run foundation models on. Systems where "usually works" doesn't count as working.
What struck me after leaving: most ops teams fight the same class of problems those systems solve, with none of the tooling. Spreadsheet reconciliations nobody owns. Ticket triage that quietly burns out a person a week. Five-plus years in machine learning taught me the models were never the gap. The gap is discipline. Pick one workflow, scope it to a dollar number, ship working code in a month.
That's what TechAegisAI is. Every engagement scoped by KPI. Every project pays for itself in ten working days or we move to a better one. Code, evals, runbooks: all of it lands in your repos, not ours.
If your team is still doing work an agent should be doing, let's talk.
Values
Three rules, no exceptions.
Outcomes over optics
Every engagement starts with a number. If the project can't be written as a before-and-after metric, it's not ready to ship.
Ship in weeks
The first workflow pays for itself in ten working days or we pause and pick a better one. The slow build is the fast build.
You own the result
Code in your repos. Models on your infra. Runbooks your team wrote. When the engagement ends, you don't need us to keep it running.
Journey
How we got here.
2019
First automation
Shipped a reconciliation bot for a retailer's ops team. Saw how much money was sitting behind tooling nobody had bothered to build.
2022
The AWS years
Three years of production ML: GuardDuty's threat detection, then Bedrock as enterprises started running foundation models on it. Function calling landed in the same window, and suddenly real agents were possible.
2025
TechAegisAI
Left AWS to put that discipline into mid-market operations. The data's messy and the stakes are real. That's the fun part.
Now
Building the playbook
Every engagement refines it. Every client takes it home. The goal is a practice that outlives any single project.
Methodology
How every engagement runs.
Audit
Two weeks with your team. Every workflow gets mapped, costed, and ranked. Output: a backlog ordered by ROI, not by hype.
Ship
Highest-ROI workflow gets built in the next four weeks. Real code in your stack. Real dashboard tracking the baseline.
Hand off
Playbook, eval harness, and runbook go to your team. We step back. Your next three workflows run without a call to us.
Ready to put AI to work?
60-minute call. No slides. We leave with a scoped plan or tell you you're not ready.