Cullom Cahill

Quality Engineer | AI Health Tech | Design Controls, V&V, Test Automation | Ex-Woebot | Open to Full-Time & Contract

As Quality Engineering Manager at Woebot Health, I took the quality function from zero to ISO 13485 certification across three design-controlled SaMD products. I built the trace matrix from scratch, developed custom automation tools for V&V, architected multi-component release workflows, and managed the full complaint-to-CAPA pipeline.Actively building AI tools into quality workflows.Open to fractional QE roles where I can combine deep quality engineering expertise with hands-on technical work in regulated AI health products.Book a 30-minute conversationSkills
ISO 13485 · IEC 62304 · Design Controls · QMS Architecture · Trace Matrices · V&V · Risk Management · CAPA · Complaint Management · Test Automation · JavaScript · Playwright · Postman · Git/GitHub · AWS · SQL · Claude Code + Playwright MCP

• C V
• Writing
• Contact

• LinkedIn
• Email

© Cullom Cahill. All rights reserved.

CV

Cahill Consulting Group LLC
Fractional Quality Engineering Leader · Nov 2025 – Present · Remote• Built an AI-powered complaint triage pipeline. 4-stage LLM workflow that classifies defects, scores probability and severity, and generates risk-ranked reports, with full compliance traceability
• Built an agentic SOP update workflow that analyzes regulatory guidance changes, generates a prioritized task list, and produces a tracked-changes Word document, reducing manual effort out of compliance documentation updates
• Consulting with early-stage software and AI product teams on QA strategy and quality system foundationsWoebot Health — AI-Powered Mental Health SaMD
Quality Engineering Manager · Aug 2023 – Mar 2025 · Remote (SF-based)• Led quality engineering function through ISO 13485 certification, built design controls processes, trace matrices, and V&V frameworks across three design-controlled products, contributed to zero audit findings
• Defined and implemented design control workflows, release processes and compliance systems adopted company-wide
• Built and maintained full requirements trace matrix from scratch, mapping V&V test coverage across multiple interconnected design-controlled components
• Trained and educated cross-functional teams (Product, Engineering, Content) on design control execution
• Architected phased release workflow with stage-specific checklists that reduced release cycle time by 3 days and reduced production deployment errors across multi-component releases (API, mobile apps, CMS, AI algorithms)
• Designed risk-based complaint and defect management system using 5x5 probability/severity matrix
• Managed cross-functional V&V execution across international time zones, coordinating simultaneous test execution
• Developed custom JavaScript validation tools that automated database comparisons across content management systems, formally validated and used in V&V processesWoebot Health
Content QA Analyst · Jul 2022 – Jul 2023 · Remote (SF-based)• Built Woebot's initial test case repository and trace matrix to support the company's first product under design control
• Contributed to integration test framework development (JavaScript/Mocha) across multiple product features
• Designed and structured test management system (Testmo) architecture and team workflowsElements Wilderness Program
Data Analyst · Oct 2020 – Jul 2022 · Huntington, UT• Lead analyst responsible for data cleaning, statistical analysis (R), and clinical outcome visualizations for the board
• Lead author on peer-reviewed journal article addressing digital media overuse in adolescentsEducationUniversity of Wisconsin–Milwaukee
Bachelor of Arts, Cum Laude · 2011
History, Sociology, and Religious Studies

• Back

Writing

Statistically Average Outputs - What Building an LLM from Scratch Actually Taught Me
LLMs aren’t reasoning, they’re predicting the next token based on the context you give them, which means weak context leads to confident but average (and often wrong) outputs. The real advantage goes to people who stay in control of the context and their own thinking.An AI Pipeline for SaMD Complaint Risk Assessment
I automated the complaint triage workflow I built at Woebot Health using a four-step LLM pipeline. Here's the architecture, the engineering decisions, and why it only works if you understand the process deeply before touching any code.
Github repoDesign Controls ≠ Paperwork Hell
Everyone assumes design control slows software down. We found the opposite. Once we built the right workflow, releases got faster, cleaner, and more predictable.Managing Multi-Component Design-Controlled Releases
Multi-component SaMD releases across time zones were a coordination nightmare. Here's the checklist system that fixed it.Bug or Complaint? How We Built a Triage System That Actually Worked
Going from hundreds of unclassified bugs to an ISO 13485-compliant complaint management system, and why building it early makes your product better, not just audit-ready.What I Learned Building a Trace Matrix from Chaos (At a SaMD Startup)
How we built a trace matrix from scratch with no existing requirements, scattered test cases, and a brand-new team, and what it taught me about traceability as a way of thinking, not just a document.

• Back