Justin Gregoire

Summary

Data scientist and AI architect. Founder of Holocron, the agentic analytics platform (MCP server with 40 tools, a 119-metric semantic layer, LLM evals, experimentation engine) that became Scribd's blueprint for enterprise AI adoption; production experience across MCP, the Anthropic API, and Claude Code. A decade defining metrics, running experiments, and turning data into executive decisions at TikTok, Peloton, Nextdoor, and Skillshare, plus client engagements at Zynga and Google Cloud Platform — causal inference and A/B testing at platform scale, from C-suite GTM plans to board-level product strategy. Also founder of Awen (awen.press), a live AI-orchestrated writing and publishing platform.

Experience

Holocron: the enterprise agentic analytics platform. One system that answers product questions, runs the experiments, and anchors how the company works with AI.

• Architected and shipped Holocron solo in 5 months: an agentic analytics platform exposing 40 MCP tools over a 119-metric semantic layer, medallion ETL, and an LLM evaluation framework — now the org's canonical interface for product questions.
• Defined the measurement foundation the company reports on: 119 governed metric definitions with typed contracts, so every team, dashboard, and agent resolves the same number the same way.
• Conducted a cross-lifecycle behavioral attribution deep dive spanning 895K trial journeys; triangulated three statistical methods and found the strongest conversion signal in the set — users who copy text convert to paid at 6.2x the odds of those who don't.
• The finding, extractors pay while explorers return, anchored executive product strategy and seeded a major new product initiative.
• Charted Scribd's shift to AI-native work: epistemic principles (outputs often wrong, verify always), AI as thought and execution partner; drove adoption through cross-functional "vibe coding" with peer review until it became how the org works.
• Built the experimentation platform end to end, from structured design intake to a Bayesian/frequentist hybrid statistical core; now the canonical surface for UGC product experiments, with 9 shipped to date.

• Analyzed user behavior across the homepage to surface Connected Fitness opportunities; recommendations shipped as a redesign.
• Led experimentation enablement across the product org: experimental design standards, tooling, and education.
• Built subscriber churn forecasts (Prophet) consumed weekly by product leadership.

• Applied causal inference to identify onboarding friction; improved new user registration by 5% at platform scale.
• Improved the logged-out landing experience (Nextdoor's largest entry point for new users) through Bayesian and frequentist A/B testing; reduced funnel dropoff by 10%.

• Benchmarked TikTok Shop's UK/SEA rollouts against Douyin in China; the recommendations, adopted by E-Commerce leadership, shaped the US launch strategy.
• Built predictive models and RFM segmentation across TikTok's user base; 5–10% higher conversion vs. baseline markets.

• Partnered with VP of Growth on international expansion; built the multi-armed-bandit pricing framework and presented the region-specific GTM strategy (pricing, marketing budget, product positioning) to the C-suite. The plan scaled Skillshare into those markets.
• Segmented teacher engagement via k-means clustering; segments anchored marketplace features that drove 5–10% of company revenue.

• Partnered with VP of User Acquisition at Zynga on multi-touch attribution (Markov chains, Shapley values); improved ROAS from 70% to 115% on $100M+ annual budget.
• Built the first real-time LTV model at Zynga; led 4-person analytics team feeding C-suite decisions on user acquisition.
• Embedded client-side at Zynga and Google Cloud Platform; ran demand-generation analytics for an enterprise developer platform, aligned departments on shared metric definitions, and delivered the executive reporting that guided marketing budget allocation.

Education

Technical

AI / Agents: LLMs, MCP server development, Claude Code, Anthropic API, agent architecture, tool use, context engineering, evaluation frameworks, retrieval-augmented generation (RAG), semantic routing, vector embeddings, sentence-transformer embeddings, prompt engineering, LangGraph, LangSmith, HuggingFace Transformers

Languages: Python (Pandas, Scikit-learn, PySpark, Pydantic), SQL, R

Experimentation / Causal Inference: Experimental design, Bayesian and frequentist A/B testing, causal inference, multi-armed bandits, hypothesis development, metric definition and measurement frameworks

ML / Statistics: XGBoost/SHAP, Markov chains, Shapley value attribution, Prophet forecasting, k-means clustering, RFM segmentation, LTV modeling, statistical modeling

Data Engineering: Databricks, medallion architecture, dbt, Apache Spark, Airflow, Snowflake, BigQuery, Redshift, AWS (S3, SageMaker)

Infrastructure: YAML-native config, Jinja2 templating, Cloudflare distribution stack, CLI tooling, Make/npm build automation

BI / Visualization: Tableau, Looker, Mixpanel, Amplitude