Product Manager · MS, HCI · MBA, Technology Strategy and Product Management (STEM)

From interfaces
to intelligence.

I'm Murtuza, a product manager with roots in digital experience and transformation. I build products from the data layer up, and I like proving they work: real models, real tests, honest numbers. Before business school, I spent nearly five years at Deloitte shaping digital product strategy and experience for enterprise clients.

See the work
  1. Interfaces

    You click, tap and type. The software waits for every instruction.

  2. Intent

    You say what you want, and the system finds what matters in your data.

  3. Intelligence

    It plans the work, runs it, and checks with you before anything important.

Work shipped with
Deloitte Tinder Wisconsin School of Business
0of test policies the AI drafted correctly in Umbra, against 55% for rules aloneUmbra · 51-case eval
0working AI products designed and built end to endOutpost · Umbra · CardRight
74usability score lift ahead of a GTM launchDeloitte
0years across product design, strategy and AITinder · Deloitte
Portfolio

Selected Work.

Products I have shaped, designed and built, from enterprise platforms to consumer apps, most of them AI-enabled. Each one starts with a user problem and ends with a working product. Open any of them for the thinking behind it.

Flagship 1 of 2 · Governance

Umbra.
Using AI to govern AI.

A product I designed and built from scratch. Umbra governs every AI tool and model a company uses: what is allowed, what data reaches it, and which agents get in. It catches sensitive data before it goes out and turns the unapproved tools it finds into a plan for where to invest in AI next. AI helps draft the rules, but fixed rules and people make the decisions.

Sample company: Kestrel Mutual, a fictional insurer with 2,400 people.

Try the interactive console ↗Read the case study ↗Code ↗
Interactive demo · sample data
Umbra at the centre · sample data Umbra console, AI map: every AI service at Kestrel Mutual on one map, with the data sources that found them Umbra console, Demand map: where approved AI would help each team most Open the interactive console ↗

Approved Unapproved High risk Centre: Umbra. Each dot: an AI tool Kestrel Mutual staff use. Each ring around a tool is a control that applies to it. Nothing turns green unless a person approves it.

22 → 0sensitive data reaching outside vendors, after one rule change
888 → 62raw alerts grouped into cases a person can act on
11 of 15AI tools staff were using had never been approved, in one sample month
Level 3 → 4on a five-level governance scale, once every agent has an owner
Demand mapShows where teams need approved AI, based on what they already use.
Corona (AI assistant)Drafts rules from plain English on a real model: 43 of 51 right in testing, against 28 for rules alone. A person approves them; fixed rules enforce them.
Agent ownersEvery agent gets a named owner, permissions and a money limit. Approved agents go to work in Outpost.
How the two flagships connect

Umbra decides what AI gets in. Outpost runs the agents that got in. Both run at Kestrel Mutual: when a person approves an agent in Umbra, it flies into Outpost carrying its owner, permissions, money limit and the approved model it runs on.

Umbra decides what AI gets in.Governance for the whole company, over weeks and quarters: every AI tool and model, what data reaches them, and which agents get in, on which approved model, with which permissions.
Outpost runs the agents that got in.Operations for one crew, in seconds: each bot acts alone up to a limit it has earned, and anything risky waits at the Gate for a person's yes.

Try it: in the Umbra console open Agents, review Agent 7f3 and choose Approve with limits. It was running on an unapproved model, so you pick an approved one. It docks in Outpost as Fern, a new bot that starts at the lowest trust level.

Flagship 2 of 2 · Operations

See every agent.
Decide the risky calls.

Outpost is a product I designed and built: the control room for the person who runs a crew of agents. Each bot acts alone up to a limit it has earned, anything risky waits for a person's yes, and every decision is logged. Here the crew looks after this site. In the full base it runs claims at Kestrel Mutual, the same fictional insurer Umbra governs.

Open the full base ↗Case study ↗Code ↗
Waking the baseOpen the full base ↗

Interactive · simulated crew. Approve a request, click a bot, send a task, or open the Autonomy lab and replay 20 shifts. It also picks up what you do on this page.

3×more work shipped when someone answers the Gate at once than when nobody does (83 vs 28 tasks a shift)
+23%output from answering in 15 seconds instead of a minute, with the rules unchanged
3× risk, +1%doubling the money limits tripled the money moving with no person watching, and barely moved output
0 of 20,000random actions got past a locked rule or a badge

Numbers from a seeded simulation with made-up work. They show how the rules behave, not how a company would perform.

The GateMoney over a bot's limit, anything sent outside and anything that cannot be undone wait for a person's yes.
The VaultOff-limits data. A bot that reaches outside its job is stopped before the tool runs. You keep it out, allow it once, or pause it.
TrustGrows with each task shipped and shows on the bot. Each level raises what it may do alone. One click takes it away.
Autonomy labChange a limit and replay 20 shifts before it is applied. You see what it buys and what it costs.
LogbookEvery decision, the rule that fired and who made it. It checks itself and exports as JSON.
One decision pointA single tested function answers allow, hold or stop. No model decides anything.

Design and Prototype Lab the interface layer of AI products.

Voice AI, fintech intelligence, behavioral products and an AI brand system, each taken from research to a working prototype. Drag, scroll, or hover to preview.

DRAG

The Human-AI Intersection.

For forty years, product design meant teaching people to operate computers. Now software acts on its own. My work sits in that shift: deciding what the AI should do, what the person should decide, and how each one knows what the other is doing.

A human hand reaching toward the AI hand

How we got here, in five interfaces.

1970sCommand line

You memorize the commands. The computer waits for exact syntax.

The humantypes exact commands
The machineexecutes, literally
Move to disturb · click to send a signal
Who does the work
Human AI
Capabilities

What I bring to a product team.

HCI training, enterprise AI delivery and digital transformation strategy, held by one person so nothing gets lost between them.

Product Strategy

At Deloitte I shaped product strategy and roadmaps for enterprise clients, including AI readiness and operating models: deciding what to build first and how it scales.

Human-AI Interaction

I started in HCI, designing interfaces people operate, from a master's in Human-Computer Interaction to product design at Tinder. Now I design for AI that acts with them, in builds like Nova, Conatus and Umbra.

0-to-1 AI Products

Umbra, Outpost and CardRight each went from an idea to a working product with a real model where it helps, fixed rules where it must not guess, and tests that prove the difference.

Research to Roadmap

JTBD, VoC and usability research turned into RICE-ranked roadmaps that shortened go-to-market cycles for enterprise SaaS products.

LLMs, RAG, tool calling
SQL, R, Amplitude
PRDs, RICE, JTBD
Figma, n8n, Jira
Murtuza

The Working Stack.

Manual workflows in, AI-native products out. Digital transformation, one workflow at a time.

The Path So Far.

Five chapters, built like a stack: human-computer interaction as the foundation, product design as the interface layer, enterprise strategy as the systems layer, business school as the decision layer, and AI product work as the intelligence layer on top.

2020
Chapter 1

M.S. (BS/MS Integrated) in Human-Computer Interaction

St. Xavier's College, University of Calcutta

Where it started: learning to design around people, from research and interaction design to prototypes that get tested.

2020graduated
2019
Chapter 2

Tinder

Product Design Intern, then Product Designer, User Experience.

Research, prototyping and redesigning the flows people use to join and start talking on Tinder.

2019to 2020
2021
Chapter 3

Deloitte

Senior Analyst, then Product Manager, Product Strategy and Digital Transformation.

Product strategy and digital transformation for enterprise clients, and shaping AI workflows.

2021to 2025
2025
Chapter 4

Wisconsin School of Business

MBA, Technology Strategy and Product Management (STEM).

May2027
  • Trusted to Lead Full ScholarshipWisconsin School of Business
  • AI PM capstoneAI-powered win/loss analytics
  • Project Manager, Graduate Business Consulting AssociationPlus Engagement Associate, Tech Product Club
  • Instructor TA, Design Innovation LabPlus Product Fabrication Fellow, Grainger Engineering Lab
Now
Chapter 5

Lighthouse AI

Senior AI Product Manager Intern.

This work is confidential, so it is not described here.

2026to now
Research & Vision

Interfaces taught people to operate computers. Intelligence will operate for them.

My research started with how interfaces shape decisions. My vision is AI that acts for people and stays accountable to them.

Human does the workAI does the work
01 · HCIDirect manipulationHuman operates. System waits.
02 · HAICollaborative intelligenceHuman asks. Model answers.
03 · AgenticOrchestrated autonomyHuman sets intent. Agents plan and act.

What I believe

  1. Intent is the new interface. People will describe outcomes, not steps.
  2. Autonomy is earned. Trust is built per task, and measured like retention.
  3. Every decision leaves a trace. If an agent acts, a person can see why.

Research

M.S. Thesis · HCI and decision-makingThe Creative Gene

How interface architecture shapes the decisions people make. The question behind everything I build now.

Read the thesis ↗

Questions I
get asked.

The short answers. For anything else, the assistant in the corner only answers from what's on this page.

murtuzabuilds@gmail.com
What kind of PM role are you looking for?

Product manager roles, especially on AI products: agents, copilots and workflow tools where the interaction model is still being worked out. I'm most useful where design, strategy and engineering need one person holding the thread.

What are you working on right now?

Finishing my MBA, interning as a Senior AI Product Manager, and leading an AI PM capstone on win/loss analytics. On the side I keep sharpening Umbra, Outpost and CardRight.

What does Human-AI Interaction mean in practice?

Deciding where the AI acts alone, where it asks first, and where it steps aside. Then designing the signals that keep a person informed and in control, like confidence, sources and an easy undo. The assistant on this page is a small example: it shows its confidence and its source, and it says so when it doesn't know.

Do you still design and build?

Yes. I started as a product designer at Tinder, I prototype in Figma, and I build my own AI tools, like a PRD-to-Jira pipeline in n8n and Nova AI, a context-aware voice assistant.

When are you available?

I graduate from the Wisconsin School of Business in May 2027 and am looking at full-time AI product roles from then.

Let's build
what's next.

Building AI products, or thinking about one? Email me directly, or pick a time for a 20-minute conversation.

murtuzabuilds@gmail.com
Month
Times are US Central.