Ved Nikolic
AI Experiences PM
Montreal, Canada
I build the experience layer where AI meets real users under real constraints. At Meta, I lead capture across AI wearable devices, including work on capture for Meta Ray-Ban Display and across Ray-Ban Meta (Gen 1 and 2), Oakley Meta HSTN, and Oakley Meta Vanguard, partnering with AI teams on ML models from ideation through production. At Best Buy, I delivered $50M+ incremental revenue by rebuilding the recommendation architecture on an ML foundation. Earlier I founded a consultancy delivering 0-to-1 solutions across fintech, blockchain, and e-commerce in four countries. I sharpen my craft through open source by building evaluation systems for AI products, a knowledge graph, automated eval loops, and adversarial analysis frameworks.
What I Build at Meta
Owned the media capture experience across Meta's AI wearable devices under real constraints: latency, battery, thermal limits. Public proof points include work on capture for Meta Ray-Ban Display and the released AI glasses line. Drove capture to the #1 feature and #1 purchase driver on Meta's AI glasses with majority user awareness and best-in-class engagement. Partnered with AI teams on ML models from ideation through production handoff. Shifted to in-house media processing, achieving sub-second latency and >99% success rate across 7+ device SKUs. Shipped the largest feature expansion in product history ahead of schedule.
How I Work
Find the real constraint
The product problem is rarely what it looks like. At Meta, the constraint was not AI quality, it was capture latency: sub-second response and >99% success rate before any AI feature mattered. At Best Buy, the constraint was not traffic, it was attach: rebuilding recommendations on ML drove a 400% increase.
Define why before how
A prototype shows what is possible. The spec defines what good looks like: which tradeoffs you accept, which you do not, and why. That clarity turned a dozen features into the largest expansion in Meta capture history, shipped ahead of schedule. Extreme clarity on the why and what is what makes the how high quality.
Stress-test before building
A/B tested every feature at Meta across wearable hardware, AI models, and companion app software. Ran recommendation experiments driving $9M+ incremental revenue at Best Buy. Built evaluation methodology proven across hundreds of LLM evals. Red-team from 12 disciplinary lenses. The gates that catch bad tradeoffs have to exist before the work starts.
Ship across boundaries
$50M+ revenue across Best Buy product lines. 3x scale across 4+ organizations at Meta. 0-to-1 products across 4 countries. 100% annual growth across 6 international markets. The hardest product problems cross team, domain, and geography lines.
Open Source Tools
Tools that solve real problems I hit as a PM. All open source, all LLM-agnostic.
Cortex
The memory layer AI coding tools ship without. Routes session context to structured destinations and surfaces cross-project patterns via a knowledge graph.
cortex save "chose connection pooling over per-request" cortex reflect # surfaces patterns across all projects
Evalgate
Evaluation methodology and tooling for AI products. Schema normalization, constraint gates, variance-aware regression detection, and cost/quality tradeoff measurement across models.
Principles from hundreds of LLM evals: - Atomic evals (one assertion per check) - Constraint gates (one failure = zero score) - LLM-as-judge variance: 5-7.5%
PM AutoResearch
Automated eval loop for product documents. Define binary pass/fail criteria, iterate, keep only improvements. Git tracks every round.
Score: 17% --> 94% (4 rounds) Evals: 19 binary criteria, locked harness
Red-Team
Adversarial analysis from 12 disciplinary lenses. Point it at any product artifact and get severity-ranked findings with grounding and worst-case scenarios.
Agents: Engineering, UXR, PMM, Privacy, Legal, Ethics, Security, Finance, Data, Design, Ops, Localization
Steelman
The blue-team to Red-Team's red-team. Takes weaknesses and converts them into positioning advantages through 6 analytical lenses.
Lenses: Strengthen weakest argument, Reframe positioning, Evidence, Expand moat, Simplify, Second-order
Stakeholder Radar
Evidence-based stakeholder profiles from real artifacts. Simulates document reviews before they happen.
Input: meeting notes, Slack threads, email chains, doc comments Output: per-reviewer predicted feedback
Career
- Drove capture to the #1 feature and #1 purchase driver on Meta's AI glasses with majority user awareness and best-in-class engagement across devices
- Partnered with AI team from model ideation through production handoff, defining performance requirements and quality metrics for ML models
- Shifted to in-house media processing, achieving sub-second latency and >99% success rate while eliminating external licensing dependency across 7+ device SKUs
- Owned privacy and compliance reviews for AI-driven features, documenting user journeys and data flows for cross-functional sign-off
- Scaled product support 3x (4 engineering managers, 40+ engineers) while shipping largest feature expansion in product history ahead of schedule
- Delivered $50M+ combined incremental annual revenue across recommendation systems and in-home service products
- Rebuilt recommendations architecture: 400% increase in attach rates, $2M+ annual revenue
- Expanded in-home services online: $30M+ expected annual value
- Founded consultancy delivering 0-to-1 solutions across fintech, blockchain, and e-commerce for clients in 4 countries
- Led discovery through launch across 6+ concurrent client engagements as sole product person
- Mentored teams across 3 time zones on product systems, data analysis, and market research
- Drove 100% annual growth across 6 global markets over 3 years
- Built 3 platforms from scratch: consumer e-commerce, gamified mobile app, custom ERP/CRM
- Increased subscription retention 60% through loyalty and gamification features
- Drove 20% year-over-year growth in mature markets through integrated marketing strategies
- Delivered product training and keynotes to audiences of 500+ nationwide
- Managed public relations, crisis communications, and brand visibility campaigns