Director of Software Engineering, Risk Tech @ Capital OneI build robots that live inside computers
Hi, I'm Madhav
Engineering executive who has spent 17 years building software, platforms, and engineering organizations across finance, healthcare, retail, and global consumer brands. These days my job is convincing machines to build the software, while convincing humans this was the plan all along.I teach computers how to listen, think, and help people. I've built things for banks, hospitals, a giant home store, and even a sneaker company!
Full name: Brameshmadhav Srinivasan. Most people just say Madhav.My full name is Brameshmadhav. That's a mouthful, so just call me Madhav!
Four things I spend my days onMy four favorite jobs
Agentic AI
Designing multi-agent systems that route work to the right specialist agent and tools, with guardrails a regulator would sign off on.
Building a team of robot helpers. Each one is great at one job, and they pass the work to whoever is best at it.
Voice AI
Built GRACE, a HIPAA-compliant voice agent for patient scheduling, with safe escalation, soft handoffs, and evaluation-first releases.
Helpers you can talk to on the phone, like a friendly voice that actually understands what you mean.
Frontend & PlatformScreens & Pipes
From Lowes.com search to patient apps: React, React Native, micro frontends, and cloud-native platforms on Google Cloud.
The buttons and screens you tap, plus the hidden pipes that carry everything to the right place, super fast.
Responsible AI
AI governance and compliance built in from day one. Member of the Responsible AI Institute.
Making sure robot helpers are fair, honest, and keep your secrets safe. Good robots follow the rules!
Brain Lab
Think first, then talkPeek inside a robot brain
People don't say every thought out loud before answering, and agents shouldn't either. My "thinker-responder" pattern: a fast, cheap System-1 model makes the silent decision (which sub-agent, which tools), then a System-2 LLM writes the reply.
Your brain has a speedy part that decides what to do, and a careful part that does the talking. My robot helpers work the same way. Pick a question and watch!
Try a customer requestAsk the robot something
Demo only. Routing here is keyword-based to illustrate the pattern.Psst: this is a pretend robot, but real ones work a lot like this.
1
System-1 thinkerThe speedy brain
Fast typed decision. Never spoken.Thinks super fast, but quietly.
waiting for a question...
2
System-2 responderThe talking brain
Writes the reply using the chosen specialist's context.Uses the right helper to give a great answer.
...
Bug Hunt
Every engineer's favorite pastimeSquash the bugs!
A 20-second break. Bugs are fastest right before a release, naturally.
In computers, a "bug" is a mistake hiding in the code. Tap as many as you can in 20 seconds!
async function helpCustomer(request) {
const intent = await thinker.decide(request);
const agent = agents[intent.route];
const tools = await mcp.connect(intent.tools);
const reply = await responder.say({ agent, tools });
return guardrails.check(reply);
}
// ship it safely, ship it often
export default helpCustomer;
Ready to debug?Ready, bug hunter?
Click the bugs before they ship to production.Bugs wiggle around. Tap them fast!
Bugs fixed
0
Time
20s
Your best
0
Design LabCode Quest
System DesignerRobot Maze
No coding required: I wrote the simulator, you make the calls. Pick the building blocks for a real launch within budget, then see if your design survives contact with users.
Coding means giving a computer step-by-step instructions. Build a program to get Bolt to the star. Plan first, then press Run!
Level 1
Bolt's program0/16
Goals
Budget used: 0 / 10 credits
The Journey
Career, one level at a timeMy video game levels
Click a level for the details.
Every job is a new level. Tap one to see what I learned there!
Years buildingYears coding
17
Since 2009, from Java trainee to directorLonger than some of you have been alive!
AI squad growthRobot team
2 → 25
Engineers, QA, and conversational designersFrom 2 builders to 25 builders
Patient NPSHappier patients
+25%
Scheduling, lab results, and billing experiencesMore people said "this helps!"
Revenue impactTeams led at once
$70M+10+
Projected annual, from 10+ teams at NikeAll around the world
Power-ups (a.k.a. the toolkit)Power-ups I've collected