MB Madhusuthanan B · Senior Staff Engineer · AI platforms and agents Manifesto
Senior Staff Engineer · Falkor

Crafting the layers below the product, for humans and agents.

I design the systems underneath the product that people and AI agents both depend on. I was part of the core team that built the Reasoning Engine, Falkor's in-house agentic framework and harness, and I now own it end to end. I also track where the agentic space is heading and bring what works into the platform.

Let's talkDown the stack ↓
Madhusuthanan B
Madhusuthanan BBITS Pilani · 13 years
Falkor
experience
13+

years building software, from fintech to industrial SaaS

live in production
2+ yrs

of AI agents running in production on the Reasoning Engine

The AI layer for the digital twinoil and gas
now exploring

LLMs meet mathematical algorithms

open to conversationsLet's build agents that hold up.

Down the stack

Now · Falkor

The AI layer for the digital twin

An oil and gas facility gets a digital twin: one live model of the whole site that people and agents can query. The Reasoning Engine is Falkor's in-house agentic framework and harness, built for that twin. It powers the Asset Copilot in the Falkor platform, and other teams across Falkor use it to build their own custom agents. I was part of the core team that built it, and I now own it end to end. It has run in production for over two years, with agents and multi-agent orchestration around one agentic loop, working with any model provider.

digital twinoil and gasagentic frameworkmulti-agentin production
From Mar 2019 · Kongsberg Digital

Condition monitoring for Kongsberg Maritime

Part of the team that built Vessel Insight, which brings data from a whole fleet to the cloud. On top of it we built applications for Kongsberg Maritime's vibration analysts: custom data-visualisation tools they used for vibration analysis in their day-to-day work, shipped both on-premise and in the cloud. As tech lead I led a team across Norway and India and owned the product's technical roadmap. I introduced behaviour-driven development, and the product was built fully BDD. Releases went from twice a year to weekly.

tech leadNorway + IndiaBDDon-prem + cloudD3.js
Feb 2018 – Feb 2019 · ItemMaster

The manufacturers portal for ItemMaster

ItemMaster was a Chicago startup with a product-content platform for consumer packaged goods. As its first employee in Bangalore, I built the manufacturers portal: the web app where brands managed their product content in ItemMaster. They could view all their products, edit product details, download product images at different resolutions, and approve, archive or reject products before they reached retailers. ItemMaster later became Syndigo, which Gladson acquired in 2019.

first Bangalore hirestartupproduct contentretail
2013 – 2018 · Wipro

Client-facing products for Charles Schwab

Four years with Wipro on Charles Schwab, across five products for Schwab's clients: a branch and consultant locator, linking outside bank accounts with multi-factor authentication, SMS alert enrolment, notification preferences and a message centre. From 2016 we built them as micro frontends, each one deploying on its own, with reusable Angular modules that other teams adopted. Test-driven development was the default.

micro frontendsAngular.NETTDD

What I can do for you

Three things I do in production, not in slides.

agents

Agents that hold up in production

Design the harness around the model: tool use, orchestration, guardrails and permissions, sandboxing, tracing and evaluations, so agents are safe to ship and easy to check.

data

Data platforms people trust

Carry data from where it is made to the people who act on it, from sensors on a ship to an analyst on shore, or from a digital twin to the agents that query it.

teams

AI-native teams without the mess

Specs before code, AI for speed, and people keeping the calls on architecture, quality and correctness.

Putting agents in front of real users or real equipment? I would like to hear about it →

Toolbox

in daily use

The Substrate Manifesto

A manifesto for AI-native engineering. Seven lines on building the layers below the product. Read it on its own page →

  1. 01

    Design for people and agents.

    Every system now has two kinds of users.
  2. 02

    Behaviour first.

    Start from the behaviour you expect, then write the code.
  3. 03

    Build the layer, not the feature.

    A good layer keeps paying off for every team that builds on it.
  4. 04

    AI for speed, people for judgement.

    AI moves fast on the rest; architecture, quality and correctness stay with people.
  5. 05

    Craft over output.

    When AI makes code cheap, knowing what good looks like, and caring enough to get there, is what's left to compete on.
  6. 06

    Ship small, ship often.

    Small, frequent releases are easier to trust and easier to fix.
  7. 07

    Own it in production.

    Building it is the start; running it is the job.

About

I'm a hands-on architect trained in software engineering at BITS Pilani. Thirteen years across fintech, product content and industrial SaaS, and I still write code. I work AI-native: specs before code, AI moving fast on the rest, and people keeping the calls that matter.

  • Kongsberg Digital → Falkor2019 – now
  • ItemMaster2018 – 2019
  • Wipro, on Charles Schwab2013 – 2018
  • Hackathon win, human-in-the-loop agentsaward
  • Founded a cross-product engineering communitycommunity
contact

Let's build agents that hold up.

Whether it is an agent harness, an industrial data platform or an AI-native team, I am happy to talk it through.

LinkedIn↗ GitHub↗ X↗ Medium↗
© 2026 Madhusuthanan B · thecompiledmindBack to top ↑
↑ ↓ to move · Enter to open · Esc to close