I own product strategy and delivery, including where AI should and shouldn’t carry the conversation, and I manage and develop a team of product managers.

See Xfinity Assistant ↗

Comcast · Xfinity Assistant

From one chatbot to a team of AI agents

Customers get help in the Xfinity app, on xfinity.com, and over messaging, with a handoff to a human agent when they need one. I lead product across that journey.

Millions

RCS messages a day

1.5M

human-assisted chats a month

Shipped

01
Expanded accessLaunched Xfinity Assistant and added RCS messaging, the rich-messaging successor to SMS, which now handles millions of messages a day.
02
Expanded capabilitiesLed the launch of bill explanations, collections, account management, and internet troubleshooting that uses customer service telemetry to check equipment and connections before answering.
03
Improved answers with generative AILed the rollout of generative retrieval in Xfinity Assistant, improving response accuracy and relevance while reducing manual content creation.
04
Reduced abandoned human-agent chats by 28%Led the launch of push notifications and asynchronous agent chat in the Xfinity app, helping customers continue conversations without staying in the app. Abandoned chats fell 28%, and average chats per customer decreased by 0.7.

Currently leading

05
Specialist agents, one interfaceMoving the assistant to specialist agents that share tools and skills, with a reusable interface serving both Xfinity and Comcast Business.
06
Human-assisted chat migrationLeading the migration of the human-assisted chat platform (1.5M chats a month), connecting AI and human support so conversations carry over between them.
07
Shopping on xfinity.comBringing the assistant into the xfinity.com shopping experience for people choosing and buying service.

Xfinity Assistant on a phone: the agent asks clarifying questions about slow internet, runs account and outage checks, then explains a detected signal issue in an alert summary

Illustrative view of the multi-agent design. Not the production architecture.

Comcast · Conversational Insights Platform

From problem to product in less than 4 months

An internal platform that shows teams what’s happening inside customer conversations. LLMs score what takes judgment, and deterministic rules calculate the rest, so those numbers are exact and repeatable.

1,000+

internal users

200K+

chat conversations analyzed daily

15+

generated metrics

01
Identified the problemTeams needed a way to evaluate customer conversations at scale and find opportunities to improve them.
02
Tested the ideaI built an AI-assisted prototype, tested it with internal users, and used their feedback to shape the MVP.
03
Built and launchedI designed the product and built the prototype that became the MVP; engineering built the production data pipeline and AI infrastructure.
04
Enabled ongoing improvementTeams use trends, metrics, summaries, and transcripts to understand issues and prioritize changes.

Conversation insights · Virtual assistantLast 14 days

TREND · SENTIMENT · DAILY

Avg +0.29

14 days agoToday

This periodPrevious 14 days

[2]How I lead and build

The principles behind my product decisions.

Start with the customer.

Understand what people need, where they struggle, and what would make their experience better. Use those needs to guide priorities and measure success.

↳ IN PRACTICE

I tested the Conversational Insights prototype with internal users and used their feedback to shape the MVP.

Reason from first principles.

Question assumptions. Break the problem down to its fundamentals, understand the constraints, and choose an approach you can explain and defend.

↳ IN PRACTICE

Voice and chat have different demands. For IVR, we prioritize model speed and cost; for complex chat, we prioritize reasoning capability.

Think big. Build for lasting impact.

Set an ambitious direction that addresses the broader problem. Make decisions that create value today and support where the product needs to go.

↳ IN PRACTICE

I led Conversational Insights from an internal prototype to a platform analyzing 200K+ conversations daily for 1,000+ internal users.

Take ownership from idea to outcome.

Connect strategy with execution. Get into the details, work through obstacles, and stay accountable for what happens after launch.

↳ IN PRACTICE

I designed, built, and launched Stillory and ImaginPaws, taking both from an idea through release on iOS and Android.

Set a high bar. Prove the quality.

Define what good looks like and test against it. Use evidence to evaluate accuracy, reliability, and the experience people actually receive.

↳ IN PRACTICE

Stillory’s bottle scanner uses repeatable evaluations. A run across 849 catalog images measured 92.3% first-choice accuracy and informed the tradeoff between accuracy and speed.

Develop people. Give them ownership.

Set clear direction, coach judgment, and give people room to make decisions. Build a team that can take responsibility and deliver with confidence.

↳ IN PRACTICE

At Comcast, I manage and develop product managers, aligning their work around customer needs and business outcomes.

[3]Built on my own

I also ship my own products: Stillory, ImaginPaws, and the agent on this page, each designed, built, and evaluated end to end.

Stillory’s Lounge home screen: bottle, tasting, and average-score stats, the top-rated pour, open bottles, and recently added bottles

Stillory scanning a bottle of Weller Antique 107 and identifying the distillery

Stillory · Designed, built, evaluated

Teaching a camera to read a whiskey label

A collection tracker and tasting journal for whiskey drinkers, live on iOS and Android. I designed and built it solo: an LLM scanner that identifies a bottle from a photo, Cooper, an AI assistant that knows your shelf, and a 1,000+ bottle eval set that every scanner change has to pass.

Visit Stillory ↗

ImaginPaws Pet Portrait screen: a photo of a tortoiseshell cat with Superhero selected and options for outfit, background, art style, and mood

The finished ImaginPaws portrait: the same cat as a masked superhero with glowing gauntlets against a lightning storm

ImaginPaws · Designed, built, evaluated

Learning to ship apps by shipping one

Turns pet photos into imaginative characters and stylized portraits, live on iOS and Android. I built it to teach myself to code and launch mobile apps, from the first screen through App Store and Google Play release.

Visit ImaginPaws ↗

Off script

Outside of work, I’m outdoors, traveling, exploring technology, or enjoying whiskey. That last one became Stillory.

Kevin on skis beside a snowboarder at the top of a run, with snowy ridges and a lake far below

Kevin in an orange jacket standing on a boulder in a rocky alpine valley under low clouds

Kevin and a companion raising a glass of wine and a cocktail outside a shingled house with a red door in autumn

On the mountain: Skiing and snowboarding all winter.
On the move: Running and biking when there’s no snow.
On the road: Traveling and exploring new places, from mountain trails to small-town inns.
In the glass: Learning what makes each whiskey different.
Always curious: AI, space, cars, autonomy, and robotics.

Where I’ve worked

Connected. Hi.

Best for roles in conversational and agentic AI. Expected wait: a couple of days, faster if you mention brisket.