Carson White
Product + Growth operator
I build products and the distribution loops that grow them.
Currently based in South Florida
ABOUT: My last 8 years
Over the last eight years, my career has moved through four places — Shanghai, Chicago, Dubai, and now South Florida — and each chapter added a capability (and a few wins) that shaped how I build today.
Shanghai was where I learned how to find signal in noise. I was doing market and channel work where the inputs were messy and the decisions mattered. I built analysis from scratch, worked across teams, and learned how to turn data into clear recommendations — the kind that leaders can actually act on.
Chicago (LinkedIn) was where I learned distribution at scale. I got deep reps in customer discovery, positioning, negotiation, and operating inside a rigorous revenue machine. I consistently performed at the top of the org — including President's Club — and saw firsthand how great products win when GTM motion is designed with discipline.
Dubai was where I learned to build end-to-end. I co-founded Prixvo, a marketplace for enthusiast and collector cars, and owned the full arc: define the product, ship it, earn trust, and drive liquidity on both sides. We generated $25M+ in bidder volume and served buyers and sellers across three continents.
Now in South Florida, those lessons converge. I'm building PoolParty — a prediction-market platform for the creator economy, designed with zero risk asymmetry, where "I'd pay to see that" becomes a real mechanism: fans commit dollars behind what they want, and creators can confidently greenlight what to produce. In parallel, I'm working on LLM-native semantic systems that make information more structured and usable. My sweet spot is the intersection of product, data, and go-to-market — taking ideas from discovery → build → launch → distribution.
Work
Featured Projects
PoolParty
Founder & Product Lead • 2025–PresentA market platform for live entertainment where "I'd pay to see that" becomes a real mechanism. PoolParty is prediction-market-like with zero risk asymmetry: fans commit dollars behind what they want, and producers/talent can greenlight what to make with clear demand behind it.
Problem
Live entertainment is a high-risk business: producers commit meaningful capital before demand is proven, and audiences have no direct way to influence what gets made beyond views and comments. The result is expensive guessing, thin margins, and lots of great ideas that never get produced.
Approach
- 1.Built an "Ideas" feed for lightweight demand discovery and rapid iteration on what audiences want next
- 2.Designed a pool mechanism that converts intent ("I'd pay to see that") into a measurable commitment without asymmetric downside
- 3.Created creator/producer workflows to review ideas, validate feasibility, and convert the best requests into funded Pools
- 4.Built engagement loops: commenting, sharing, and social proof to help demand aggregate around specific moments
- 5.Implemented clear thresholds and fulfillment mechanics so creators can confidently greenlight when demand is real
- 6.Partnered with a launch creator-sports league to pilot real pools around matchups / programming decisions and iterate quickly
Outcomes
- •Shipped a functioning V1 with Ideas feed + Pool pages + social engagement primitives
- •Secured a launch partner in a creator sports league (PPV-oriented) to run initial market tests
- •Established a repeatable structure for turning audience requests into producible events (and a pipeline of candidate Pools)
- •Built foundational GTM narrative: "Fans don't just watch — they decide."
Carvia.ai
Co-Founder & Product / Growth Lead • 2024–PresentAn automotive intelligence layer for voice AI and agent companies. We create both semantic (LLM-readable) and heuristic (rule/score-based) vehicle data so agents can understand a car's story, interpret risk, and respond with decision-grade confidence.
Problem
Voice AI and agent companies are scaling quickly, but in automotive their agents hit a ceiling: they can converse, but they don't understand the domain deeply enough. Automotive decisions require precise context — history signals, title/liens, market pricing, ownership patterns, and risk indicators — yet the available corpora are either shallow, inconsistent, or trapped in static report formats.
The gap isn't the model. It's the understanding layer: without structured semantics and reliable heuristics, agents become brittle, miss edge cases, and lose trust in high-stakes workflows.
Approach
- 1.Normalize the raw world: aggregate fragmented vehicle signals and standardize them into a consistent event + attribute schema
- 2.Add heuristics: derive decision-oriented features (risk flags, scores, and "what matters" summaries) that perform reliably across edge cases
- 3.Add semantics: generate LLM-friendly representations so agents can interpret the vehicle's story and answer natural-language questions accurately
- 4.Design for retrieval: package the corpus into retrieval-ready chunks + structured outputs to reduce ambiguity and hallucination risk
- 5.Integrate into agent workflows: expose the data in partner-friendly ways (embedded components, APIs, agent tool calls) for real-time use
- 6.Close the loop: instrument coverage/quality feedback so the corpus improves with every query and partner interaction
Outcomes
- •Built a dual-layer dataset (semantic + heuristic) specifically designed for agent understanding, not PDF-style reporting
- •Demonstrated early traction for this wedge with voice/agent teams looking for deeper corpora such as Feather (YC S22), Podium, & Matador.
- •Established a repeatable pipeline for turning fragmented records into consistent, decision-grade agent context
- •Created a foundation that supports multiple surfaces: agent integrations, partner widgets, and consumer-facing report experiences
Prixvo
Co-Founder & Product / Growth Lead • 2023–2025Co-founded Prixvo, a digital auction marketplace for enthusiast and collector cars built for the GCC. I led product and growth end-to-end — building the platform, driving liquidity on both sides, and earning trust in a high-friction category.
Problem
The GCC used/collector car market is huge, but trust is thin. Shoppers face opaque pricing, misrepresentation, and incomplete vehicle information — while sellers struggle to reach qualified buyers without dealer markups and games. The result: high intent, low confidence, and transactions that don't clear efficiently.
Approach
- 1.Built an auction-first marketplace experience designed for transparency, urgency, and qualified demand
- 2.Structured the supply playbook (seller onboarding, listings standards, inspection detail expectations) to raise trust and conversion
- 3.Developed the demand engine: content, community hooks, and distribution partnerships to bring bidders to the platform
- 4.Implemented high-signal vehicle presentation (details, context, and disclosure) to reduce uncertainty and increase bid confidence
- 5.Designed a clean transaction workflow and policies to reduce friction and protect both sides
- 6.Measured everything — optimizing listing quality, bidder activation, and close rates through tight iteration loops
Outcomes
- •Generated $25M+ in bidder volume across listings on the platform
- •Reached a global buyer base, with participation spanning three continents
- •Built a repeatable marketplace operating system: supply acquisition → listing quality → bidder activation → conversion
- •Established the brand as a trusted auction partner in a region where trust is the core constraint
- •Exited (No — not a life changing sum of money)
Plex
Author & Product Lead • Internal ToolPlex is an internal system I built to support other projects — designed to turn multimedia content (video, audio, long-form pages) into annotated, structured, machine-usable knowledge for retrieval, agent workflows, and dataset development.
Problem
Across multiple builds, the hardest knowledge to reuse wasn't text — it was multimedia. The most valuable insights live inside podcasts, interviews, panels, and long videos, but they're locked in timelines. Even with transcripts, it's hard to search, cite, and reuse the real meaning of what was said.
Standard retrieval is shallow: it finds similar phrases, not reliable understanding. What was missing was an annotation layer that could extract entities, claims, relationships, and context — and tie them back to the exact moments they came from.
Approach
- 1.Defined a canonical annotation schema for multimedia (entities, claims, relationships, topics, timestamps, provenance)
- 2.Built a static pipeline: transcription → segmentation → normalization → annotation
- 3.Implemented semantic annotation outputs so content could be queried by meaning, not just text similarity
- 4.Produced retrieval-ready artifacts with timestamped references and predictable structured fields for downstream agents
- 5.Added lightweight heuristic scoring hooks (source type, consistency checks, citation density) to rank trust and usefulness
- 6.Documented the workflow and packaging format so other projects could reuse the pipeline without rework
Outcomes
- •Delivered a working static pipeline that converts multimedia into annotated, structured knowledge artifacts
- •Made long-form audio/video searchable and referencable at the "exact moment" level via timestamped provenance
- •Established a reusable internal pattern combining semantic annotation + lightweight heuristics for more reliable retrieval from multimedia
Current Focus
Now
Building a new product in the creator economy space. Exploring how AI can enhance human creativity rather than replace it.
Previously: Led growth at a B2B SaaS company, built marketplace platforms, and shipped data products that scaled to thousands of users.