Thomas Howard

Promoted to CTO of this 10-person healthcare SaaS company, which makes referral and Annual Wellness Visit (AWV) software for primary care clinics, after four years building its AI platform; own Applied AI strategy, product architecture, and engineering delivery.

• Lead Sally, an LLM-powered healthcare voice agent live in production at pilot scale, running patient outreach calls for primary care clinics: real-time browser and phone conversation with interruption handling, tool execution, background analysis, and structured visit-brief extraction.
• Designed the boundary between model behavior and deterministic consent, identity verification, authorization, and tool access in patient-facing workflows.
• Built scenario-based evaluation harnesses covering conversational behavior, structured outputs, safety rules, and failure handling as part of Sally's production hardening.
• Architected Symphony, coordinating 10 specialized engineering agents across implementation, testing, review, debugging, and remediation with isolated execution, signed leases, tool permissions, cost accounting, and audit trails; humans approve every merge and deploy.
• Built four applications in four months on my own with Symphony, across enterprise healthcare, workflow automation, and patient engagement; live in production at pilot scale, with ~1,200 users across 4 primary care clinics and no production incidents to date.
• Set the Applied AI roadmap with the CEO and Product; hired and led a team of three engineers and now lead a senior engineer, training him in AI-assisted development, while staying hands-on in Python, TypeScript, APIs, and healthcare integrations.

• Led LLM reliability on Amazon Bedrock, building evaluation workflows, goldset reference datasets, and guardrails.
• Designed an AI engineering control plane: specialized monitoring, security, and SRE agents producing structured evidence, with deterministic triage, bounded remediation, protected verification gates, and human escalation.
• Built an AI test-intelligence system that decomposed product stories, generated test requirements and API tests, analyzed defects, and targeted retests.
• Architected an explainable physician-referral ranking engine in Python and scikit-learn, combining clinical quality, outcomes, eligibility, availability, and proximity with deterministic scoring, uncertainty handling, and abstention.
• Developing supervised models to predict 30-day referral completion, with leakage prevention and evaluation on precision, recall, ROC-AUC, and calibration.

• Built and operate a production consumer SaaS end to end: React/TypeScript and Node.js/Express with REST APIs, PostgreSQL, and Redis.
• Use Codex and supervised agent workflows for implementation, test generation, debugging, and review, with CI verification and human-controlled releases.
• Designed idempotency, retries, data-correction paths, rate limits, and provider budgets so the product degrades gracefully under failure and keeps operating costs in check.
• Run the AWS runtime on ECS Fargate, migrated from Kubernetes to fit the product’s scale, with VPC, IAM, CloudFront, and S3 defined in Terraform, and deployments with health checks and rollback.

• Built and led a 12-person engineering organization at this startup, through an engineering manager, owning hiring, technical direction, mentoring, performance management, and leadership development.
• Reduced regression cycle time by approximately 90% through an automation-first transformation, earlier validation, and repeatable release controls.
• Directed CI/CD, developer tooling, AWS architecture (EC2, RDS, ECS), and production observability across multiple product lines with product and engineering leadership.