Speaker

Raj Kumar

Raj Kumar

Rails Tech Lead at Sycamore – putting LLM agents on regulated production Rails.

Toronto, Canada

Actions

Tech Lead at Sycamore Informatics, shipping production LLM agents across an 8-product Rails ecosystem for regulated clinical-trial data. Sycamore is the market leader in clinical trial data science, serving 10 of the Top 20 Global Pharma — AbbVie, BMS, Bayer, Daiichi-Sankyo, CSL, Genmab, Alcon. Scale: 2,000+ TB, 10,000+ active users, 100+ regulatory submissions.

15+ years across Rails platforms, AI-driven engineering, and technical leadership. Former Director of Engineering at Safyre Labs, where he ran the multi-brand B2C ecommerce platform behind Jewlr, Lee Fiori, and ONE+ONLY (1M+ customers). Earlier: AI-driven initiatives at Mars, Inc., Etsy, and IKEA; Rails/AWS delivery for Bossa Nova Robotics, HackerRank, Agero, etc.

Currently focused on the intersection of LLM agents and regulated production Rails — fails-closed customer-environment resolvers, idiom-aware skill-layer invariants, version-pinned code reads. SOC 2 Type II, ISO 27001, HIPAA, ICH-GCP, CDISC, FDA/EMA/PMDA-aligned.

International conference speaker & tech paper author. Recent: PHUSE EU Connect 2025, PHUSE SDE Shanghai 2025, CDISC US 2025, Toronto Ruby 2024, Toronto DevOps Exchange 2023.

Area of Expertise

  • Health & Medical
  • Information & Communications Technology
  • Law & Regulation
  • Physical & Life Sciences

Topics

  • ruby on rails
  • ruby
  • Artificial Intelligence
  • LLMs
  • Agents
  • Architecture
  • AWS
  • MCP
  • Claude
  • AI Agents in Production
  • Regulatory Compliance
  • Clinical Trials
  • LLM Observability
  • Developer Experience
  • AI Security

Rails Idioms That Break Your AI Agent

You asked your LLM agent to speed up a folder-move operation in Rails. It writes the textbook idiomatic answer. A whole folder tree goes invisible to its users the next day.

Not a Rails bug. Not a Claude bug. Rails and Claude each did the idiomatic thing, and the idioms collided.

I'll walk three real production incidents from our regulated clinical-trials Rails fleet where this happened – around ActiveRecord callback chains, AR-backed job queues treated as ordinary ActiveRecord tables, and Rails's transaction-by-default. Each trap fires specifically because of the Rails idiom; the convention chain IS the failure surface.

Side-by-side: our actual Rails source, the agent's actual first-pass draft, the skill-layer invariant that catches it before it ships.

Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.

Jump to top