1 Who I am

I build AI products, from research to revenue

For the last 4 years, I’ve been building platform, API, and AI products across consumer, enterprise, and developer markets: from speech and LLM products at a frontier AI lab to no-code computer vision on the factory floor.

Get in touch → See the AI work
Meg Ross, smiling, in black and clear square glasses and a plaid shirt
fig. 01: Meg Ross, M.A.
2 The proof

AI that works outside the demo.

Evaluation, data quality, testing, and adoption: the unglamorous work that gets AI products to real users and real revenue.

~30h Saved every week

A custom automated LLM benchmarking and evaluation scoreboard, architected with AI researchers and built in Claude Code.

Subquadratic
2 mo To MLR approval

Near-zero hallucination rate, measured against a 2,000+ question golden evaluation bank, through HITL validation.

Re:Course AI
~100k Hours of training data recovered

Found and fixed metadata and transcription defects in golden datasets.

Subquadratic
75% Less debugging time

Estimated, from an automated behavior-driven testing suite for the TTS API.

Subquadratic
+33% GenAI tool adoption

30-day retention lift across 129 PMs, coaching AI-native product-led growth.

Adobe
3 mo To ship no-code ML onboarding

Computer vision for non-technical manufacturing and security teams, opening up a ~$3B market.

Groundlight AI
$120K ARR from a converted pilot

A Vanderbilt University pilot turned annual contract, with pharma contracts worth $750K ARR in the pipeline.

Re:Course AI

The AI toolkit

Evaluation

LLM benchmarking (NIAH, RULER, SWE-bench Verified) · LLM-as-judge · HITL validation · golden datasets · bias evaluation

Models & modalities

Model fine-tuning · conversational GenAI · speech AI (TTS, STT, ASR) · computer vision

Building

Claude Code · Cursor · Python · SQL · JavaScript · AWS · Google Cloud · Kubernetes · Vercel · Cloudflare

3 Is this you?

AI problems I’ve already solved

If one of these sounds familiar, let’s talk.

EvaluationYour team validates models by hand.
What I did

Architected an automated LLM benchmarking and evaluation scoreboard with AI researchers in Claude Code, saving ~30 hours of engineering and evaluation time per week.

Subquadratic
Safety & complianceGenAI output has to clear a regulatory review.
What I did

Designed and ran HITL validation across multiple fine-tuning cycles, personally reviewing model outputs, to reach a near-zero hallucination rate on a 2,000+ question golden evaluation bank and MLR approval in two months.

Re:Course AI
UsabilityYour ML product only works for engineers.
What I did

Turned engineer-centric computer vision tooling into a no-code onboarding interface for non-technical users in three months, across 10 engineers and 4 ML scientists.

Groundlight AI
Data qualityYour training data has silent defects.
What I did

Recovered ~100k hours of training data by identifying and fixing metadata and transcription defects in golden datasets.

Subquadratic
ReliabilityDebugging your AI API eats every sprint.
What I did

Wrote an automated behavior-driven testing suite for a TTS API, cutting debugging time by an estimated 75%.

Subquadratic
AdoptionYou rolled out an AI tool and nobody uses it.
What I did

Lifted 30-day adoption of an internal GenAI productivity tool by 33% across 129 PMs by coaching AI-native product-led growth workflows.

Adobe
4 Let’s talk

Let’s build something.

Next, I’m looking for a long-term home: a team scaling a product people already pay for.

01

Hiring a product leader?

I’d love to hear about the role and the team. Send it my way and let’s talk.

02

Want to talk shop?

Product, learning science, or building things with Claude Code. Happy to compare notes.

LinkedIn ↗