The fractional CTO your AI company is searching for. The best fractional CTO for AI startups, with a real track record. A senior AI consultant for seed to Series B teams. A top AI architect for hire, with agentic workflows, advanced RAG and low-latency voice in production. Fractional CTO for AI companies, Chief AI Architect at TWINii. Your part-time CTO, six years of AI in production. Technical due diligence from someone who ships AI

Where AI stops being a feature and becomes a system.

Agentic workflows, advanced RAG with context engineering and low-latency voice, with the evals, guardrails, LLMOps and token economics that keep them standing once real users arrive.

Book a 20-minute call hello@msingh.com

6 years in production AI16 years shippingNow building TWINii

01What I work on

Agentic workflows, advanced RAG, low-latency voice and the evals under them.

Watch one request go through a system I would build. Every layer is a place where AI products break, and where I spend my time.

CustomerCan you move my Thursday session to four?

Assistant

Latency: 0 ms
Cost: …
Eval: running

Hover or tap any layer to see what it does.

Eval harness

Every release tested against real traffic, in every language you serve, before it clears the release gate.

LLMOps in 22 languages at Devnagri

LLMOps

Every step traced: latency, tokens and why it failed. Model routing on the evidence.

Observability at TWINii and Devnagri

Token economics

Cost per resolved task, with retries and human follow-up counted in.

Cost control at TWINii

02Track record

Self-taught, since 2010.

I learned by shipping: first for clients around the world, then for startups that had to scale. The last six years, production AI that had to hold: agents, retrieval, voice.

See the whole track record

First

Started on a phone

I wrote my first code on an LG Cookie, in the quiet hours in my own store. A computer came soon after, and with it 350+ freelance projects through Envato Studio, Elto, Fiverr and CodeCanyon.

Then

Products that had to scale

Technical lead, core product
Built the core integrations architecture. It went on to become an Indian unicorn.
Software architect → acting CTO
Built the team, and the first product went live in 45 days.
Solutions architect
Refactored a five-year-old legacy system and made it multilingual.
UI/UX architect
Modernised the front end of its business tools.
Solutions architect, five years
A Silicon Valley B2B marketplace: real-time data, integrations, cloud-native.

Now

Six years of production AI that had to hold at scale

Devnagri AIArchitect, then fractional CTO

LLM workflows, voice and RAG for a localisation platform across 22 languages.

LLM workflows
LLMOps
Voice
RAG
22 languages

TWINiiChief AI Architect

Real-time AI twins of real experts. Ten went live at launch.

Low-latency voice
Cloned voice
Advanced RAG
Memory
Guardrails
MCP tool layers

And AI products of my own: onii.one and an AI-first email service.

Teams I have built with, plus 350+ freelance projects

03In their words

People I have built with.

Founders and engineers who have shipped with me, in their own words.

“Mandeep didn't just put in a patch, he helped redesign the system architecture to handle 5x the load without too much of extra server hardware.”

Jaspreet Singh OberoiHead of Engineering, Devnagri

“Mandeep is the level-headed, methodical thinker you want on your team. He is capable of looking at the big picture, anticipating problems and planning before hand.”

Rajat HansCo-founder and CTO, Kenko AI

“The guy thinks 5 steps ahead. He bridges DevOps rigor with AI workflows like it's second nature. Clean infra, predictable deploys, zero drama.”

Kamaljit SinghFounder, IndraOps Technologies

“It's like he can see inside your head what you want to explain and will start improvising on them, and then all you will be saying is yes, yes, exactly.”

Kumar RaviAssociate Staff Engineer, Nagarro

Ways I can help you.

Ongoing

On your team

as a Fractional CTO, an AI Architect, a Head of AI, an AI Solutions Architect

I shape the architecture and take the model selection and vendor calls with you, in writing, with the reasoning.
I own delivery: what ships, when, and whether it clears the bar before customers see it.
I own quality and security: the eval harnesses, the programmatic guardrails and the answers your enterprise buyers ask for.
I own the token economics: cost per resolved task, by model, by step and by retry.
I keep a fixed slice of every week for your team, stated before it starts, and a line to me between calls.
I leave a decision log your board can read, so the reasoning outlives the call.

Call it what fits your team. The job is the same: the hard calls, taken with you, and a team that knows where it is going.

As needed

A one-time engagement

For one specific need, with a clear end. In the order teams usually need them.

01Demo-to-production plan to take your system from a working demo into production, with the model and vendor calls and the reasoning.
02Decision session on the one call holding up the quarter: a model, a vendor, build or buy. A decision you can defend, written up.
03Architecture review before you scale: what gets expensive or fragile at ten times today’s usage, and what to change first.
04Production readiness audit of your agent on your real use cases, against the six gates a production system has to pass.
05Release gate that every release has to clear before customers see it: evals, guardrails, regression set, rollback.
06Enterprise buyer readiness pack with the answers your buyers will accept on security, reliability, data handling and cost.
07Compute cost reduction pass to find where the money goes per resolved task, and what to change first to bring it down.
08Technical due diligence on an AI company you are about to back, before the wire. Only where I have no stake.

What I build when the day is done.

More work and products

onii.oneLive in alphaShip the campaign the day you think of it. Describe the offer. Get a published campaign page that captures leads, on your own domain, in minutes.

SidekickComing to the Mac App StoreGoogle Tasks, finally on your Mac. Your list floats over whatever you are working on. Ticked off here, done on your phone.

In progress

An AI-first email service. Email that reads, sorts and drafts with you, with agents, tool calls and model routing under the hood. Early days.

For my kids

Sukhi Play and Jazz’s Studio. A locked-down browser for my son and a stationery studio for my daughter. Free for anyone.

Open source

polinrider-cleaner. Finds and cleans up after the PolinRider supply-chain campaign. One binary, dry-run by default, actively maintained.

06Writing and code

I write about AI systems in production.

Notes from the work as it happens: eval harnesses, context engineering, token economics, low-latency voice, and the code behind some of it.

LinkedInWhat AI systems in production teach me: eval harnesses, context engineering, token economics, low-latency voice.

XQuick notes on agentic workflows, LLMOps and running production AI, as they come.

GitHubThe code: open-source tools, experiments and products.