Description

Hello, I am Enrico Teotti. I help software teams integrate AI into production systems, and build the product discipline to keep it working.

Over the last year, I rebuilt an LLM classification pipeline from 68% recall to 95% using an ensemble of heuristics, ML, and a multi-step LLM chain (without increasing API costs). I cut a 32-second p95 response time to 600ms by connecting an AI agent to Sentry, Posthog, and a production database. I course-corrected an 80-prompt LLM pipeline that shipped with no evals, where operators were discarding most of the AI output.

Before that: 20+ years of software engineering and product management. Principal Engineer at Pivotal Labs (Java/Spring, React, Ruby on Rails). Senior Product Manager at Artium. Capacity planning to 300K req/min. Products serving 4M+ users. I've been writing code since 2001 and managing products since 2017.

What I do now: AI/LLM integration strategy, production ML pipeline architecture, product strategy and roadmap prioritization, hands-on engineering on legacy and greenfield codebases, and workshop facilitation for product leadership alignment.

I write about all of this on my blog

Industry field of expertise

Languages

English
Native or bilingual

Workplace preferences

Remote only

Primarily works remotely

teotti.com
Independent Consultant - Product & Engineering Leadership
January 2001 - Today (25 years and 5 months)
- Built an ensemble classifier (heuristics + TF-IDF ML + multi-step LLM chain) that raised recall from 0.69 to 0.95 and F1 from 0.78 to 0.90 without increasing LLM spend. Human review dropped from ~1,000 jobs/day to ~90 - cutting a domain expert's daily QA from 5 hours to 30 minutes.
- Inherited an 80-prompt LLM pipeline for extracting structured data from contracts - no evals, no audit trail, no dashboards. Introduced per-attribute error measurement, redesigned prompt architecture for safe iteration, replaced a failing AI QA step with deterministic safeguards, and built the eval suite that gave the team a baseline.
- Used an LLM agent connected to Sentry, Posthog, and a production database replica to diagnose a 32-second p95 response time. Confirmed the responsible feature had 2 clicks in 30 days. Shipped a fix that dropped p95 to 600ms.
artificial intelligence Test driven development
Intercollegiate
Senior Product Manager
January 2025 - March 2026 (1 year and 2 months)
Europe
• • Defined company-wide standards for AI reliability - evaluation frameworks, test datasets, prompt version control, cost controls, and human-in-the-loop benchmarks.
• • Built an MCP server for analysts to query production data using domain language. Coached non-technical staff to build internal tools and run autonomous discovery with Claude Code.
• • Used event storming to map company processes and surface bottlenecks and pain points with data.
• • Kicked off and delivered 5 projects across engineering and data teams.
Artium
Senior Product Manager
June 2021 - June 2024 (3 years)
Tampa, FL, USA
• • Led product integration of alternative investments into a B2B wealth platform for a US custodian with $44.3T AUM. Aligned cross-division objectives, built strategic roadmap, and provided senior leadership with weekly progress updates.
• • Delivered portfolio analytics (performance, risk, attribution, shock testing) one month early. Used Python/Jupyter to validate algorithms and unified Figma designs across siloed teams.
• • Led agile transformation at a Fortune 50 logistics company (~600K employees). Facilitated story writing workshops that reduced story rejections by 70% and defects by 80%.
• • Guided product strategy for a B2C fashion marketplace (4M users). Built user research systems and led data-driven experiments, including rolling back underperforming features to save resources.