Mike Aristocrat · AI Product Manager · New York City
From computer vision in 2017 to agentic AI at DoorDash today.
Software I built has shipped inside
From single features to platforms and multi-product suites: every era shipped to real users, with the data flywheels and evals that kept them working.
AI for a global marketplace, across chat and voice. Built a 0→1 co-pilot augmenting support agents on the most complex cases: cut case handle time 8% while lifting communication quality 10%. Now scaling voice and chat AI to consumers across all three major DoorDash brands: DoorDash, Wolt, and Deliveroo.
Invented Sidd Spark (patent pending), a GenAI contact-center suite, pitched and sold alongside the C-suite. Built the evals (NER, intent, transcription WER) that made models swappable, and the operational engine that scaled it from zero to 70+ enterprise brands in 15 months, reengineering the product so a deployment took under a week instead of three months.
Defined and launched Symon.AI, a no-code data & autoML platform that let analysts build ML without engineers, removing the technical barrier to AI in a $1B+ market, pre-generative-AI. Personally landed Shopify as the first customer, building executive dashboards used at the very top of the org.
Built a 0→1 AWS ML platform unifying IoT data from 600k+ vehicles; launched the company's first end-to-end ML product, fuel forecasting; and shipped an external model-inference pipeline for real-time predictive maintenance.
Pioneered an AI dermatology app, full product lifecycle: built the world's largest real-world skin-disease dataset (0 → 500k+ images) and a 50+ dermatologist labeling network, ran clinical benchmarks against physicians and dermatologists, and navigated a dynamic regulatory environment to the FDA's door.
From 12 LinkedIn recommendations: managers, peers, and direct reports across four companies, every year from 2019 to 2026.
"Quickly established the gold standard for a true Product Management professional… spearheaded a truly transformative, landmark product… fundamentally changing our business trajectory."
Jeff SaundersVP of Software Engineering, Laivly"One of the best product partners I've had… he never loses sight of impact. What stood out most was how close he stays to the customer."
Carey RansoneDirector, Strategy & Operations · worked together at DoorDash"Relentless about understanding and breaking down customer problems no matter how ambiguous… exemplifies failing fast and learning quickly. He knows when he needs to lead from the front and when to lead from the back."
Faris HijaziHis manager at Fleet Complete · now Product Lead at Google"Working under Mike's leadership was the period of my steepest career & personal growth… he launched zero-to-one products and drove the organization toward a product-led model."
Victoria PaskannayaProduct Manager · reported to Mike"One of the most effective Product Managers I've ever worked with… his support for product design is particularly noteworthy: he collaborates closely with designers, offering thoughtful feedback and fostering an environment where creativity thrives."
Shine ChaudhuriLead Product Designer"One of the most effective product leaders I've collaborated with… a rare combination of strategic vision and practical execution. He's the kind of leader who makes everyone around him more effective."
David SheridanVP of Sales / Country Lead Canada, Laivly
"A rare combination of confidence and humility… he'll argue his points firmly and with evidence, but change his mind the moment he learns something that warrants it. His ability to rapidly gather evidence, design, build, test and iterate helped us launch our most successful product yet."
"A driving force behind our GenAI and LLM initiatives… his leadership in orchestrating our most extensive GenAI data annotation efforts enabled the creation of high-quality datasets, which became the backbone of our deployed AI solutions."
"Under his leadership, the data platform team became an ownership-driven team able to deliver a cloud-based platform enabling adaptability and efficiency for the organization… an excellent product and engineering leader."
"A true generalist and growth-minded individual, with a relentless desire to understand, reflect, and learn… his combination of strong business acumen with data literacy and customer focus makes him an invaluable asset for any team building products in data and technology."
"A leader inspiring excellence from those around him… able to coordinate projects amongst different departments and nimbly adjust to ever-changing situations."
"A true project owner that ruthlessly prioritizes end-to-end processes… he clearly articulates what needs to get done, why it needs to get done, and trusts his team to get those things done."
"Just deploy AI" isn't a strategy. These are the unglamorous things that actually make AI products work. I've built each of them multiple times.
Live in the data. Touch the outputs. Read the transcripts until your brain is numb: LLMs will tell you about 33% of what you need to know, and the rest you can only infer yourself. Then build evals. They're your friend, and they're worth it almost every time. Build the first few by hand, then scale with an LLM. That's how real performance comes.
The lesson that keeps repeating: you can't simulate or assume an eval set. The performance you measure will be misleading. An eval has to reflect the real conditions the AI will work under; that's what gives it a decent start. Clinical images from doctors and photos taken on a consumer's phone are radically different distributions. When people want an AI to step into a conversation can't be invented at a desk; you have to understand how they actually think and type. And contact-center policy says one thing while agents handle cases another way, and your eval has to bridge that gap.
Data strategy is its own competency: getting the data, cleaning it, producing it, organizing it, structuring it. Data moves through phases over a product's life, and what you can build at any moment depends on what your data can support. Early on, that usually means bootstrapping it yourself until the product can improve on its own.
Feedback loops aren't magic. Today they take an enormous amount of human review, and I don't diminish that effort: I embrace it, doing the review myself until the value is clear. Then I build the machinery that scales it (dermatologist networks, Mechanical Turk, classifier-assisted review, automated QA) so ML integrity stays high and the improvement loop is durable.
Latency budgets, cost-per-interaction economics, hallucination controls, enterprise-grade SLAs. And the one that matters most: no one cares that there's AI behind it. It's gotta actually provide value.
Independently checkable: awards, filings, and published research.
The same playbook, proven across five eras and businesses: data flywheels, evals, humans in the loop, and relentless shipping.
I find the highest-leverage AI application in a business, then ship it with the evals and data loops that keep it working after launch. I've pitched, demoed, and sold to C-suites, and built products CFOs and CROs used weekly. That's how a GenAI concept became a suite inside 70+ enterprise brands in 15 months.
Ask about advisory →Product strategy, evals design, data annotation ops, pipeline architecture, jumping on customer calls, writing the FDA submission: I've done every job adjacent to the model. The zero-to-one phase, where nothing is defined yet, is where I do my best work.
See how →Scoping your company's highest-leverage AI initiative, building something ambitious, or working on the frontier? I'm in New York, and I'd love to meet you.
Mike Aristocrat is an AI product manager and product leader based in New York City. Over nine years he has built AI products end to end across five technology eras: computer vision at Triage, big data and ML platforms at Fleet Complete, no-code autoML at Varicent, generative AI at Laivly, and agentic AI at DoorDash, where he leads voice and chat AI across DoorDash, Wolt, and Deliveroo. Products he built have shipped inside 70+ enterprise brands and reached more than 10 million consumers.