# Patrick Kelly AI-Native Platform Engineer | Product Architect | Backend Systems [City, State] | [email] | [phone] | [LinkedIn] | [GitHub / portfolio] ## Live Resume Agent For hiring-manager and recruiting-agent review, this resume includes a gated live portfolio agent: - URL: `[portfolio URL]/?src=openai-principal-b2b&access=PK-OAI-B2B-2026#chat` - Access code: `PK-OAI-B2B-2026` - Agent-readable manifest: `[portfolio URL]/.well-known/agent-resume.json` The live agent answers from a public-safe source brief and does not store conversation transcripts by default. It logs only limited source/topic metadata for abuse control and to understand which submitted resume was reviewed. ## Summary I build AI-native business systems where product judgment, backend architecture, data, and operations all have to meet in the real world. My strongest work has been turning ambiguous, regulated workflows into production software: designing APIs and data models, shipping FastAPI/Postgres/Next.js systems on GCP, building RBAC and compliance controls, integrating messy operational data, and using AI agents as leverage without giving up engineering judgment. I also have a long-running pattern of building toward the next interface before the market has convenient vocabulary for it: automating B2B workflows, designing delegated production tools, and treating software as an active collaborator before "agent" became the common label. I am most useful in the space between "this should exist" and "this is running safely in production." ## Core Fit For OpenAI B2B Engineering - Backend and platform architecture for production business applications - API design, domain modeling, database migrations, and service-layer ownership - Secure enterprise workflows: auth, RBAC, auditability, HIPAA/FERPA-aware data handling - AI-native internal tooling, agent orchestration, embeddings, search, and developer leverage - Future-facing automation and workflow design before categories become obvious - Product-minded engineering across clinical, finance, HR, operations, and executive workflows - Comfortable turning ambiguity into specs, systems, tests, launches, and operational fixes ## Experience ### Lighthouse Therapy - BrightBot Platform Head of Development / Head of AI Operations & Strategy [Start date] - Present - Led architecture and hands-on buildout of BrightBot, a healthcare and education operations platform for Lighthouse Therapy spanning EMR workflows, billing, scheduling, therapist/coordinator portals, executive dashboards, HR tools, customer intelligence, and AI-assisted operational support. - Designed and shipped production backend systems using Python, FastAPI, SQLAlchemy, Alembic, Pydantic, PostgreSQL, pgvector, and role-scoped service layers, with a Next.js/TypeScript frontend and GCP Cloud Run / Cloud SQL deployment model. - Built secure enterprise foundations including Google OAuth, NextAuth, JWT-based backend auth, Google Workspace group/OU role mapping, RBAC scopes, soft deletes, audit-oriented workflows, and HIPAA/FERPA-aware data handling. - Designed API and data contracts across EMR, finance, CRM, HR, marketing, and clinical workflows, using schemas and service boundaries to keep operational complexity from leaking into the UI. - Owned production data migration and reconciliation from Monday.com into the EMR, including ESY 25-26 ingestion: 112 active students, 121 IEP services, 46 therapists, 6 coordinators, 15 school contracts, and lossless raw-source archival for future auditability. - Root-caused and remediated production blockers across data, migrations, and workflow logic, including assessment-type seed corruption, missing EMRBase audit columns, appointment lifecycle bypasses, RBAC visibility gaps, and Alembic multi-head deploy failures. - Built the Customer Intelligence Engine with pgvector-backed embeddings, hybrid search, HubSpot sync, customer profiles, meeting-prep and role-play services, 13 API endpoints, 6 agent tools, seed/embed scripts, and 24 passing tests including compliance checks. - Created implementation specs for eight department tools, including shared component patterns, FastAPI endpoints, database schema, RBAC scoping, data flow, UI/UX, and integration points. - Helped move the organization from scattered tools and manual operational knowledge toward a coherent platform with auditable data, clear workflows, and AI-assisted execution. ### DevBrain / Nooma Stack Founder / Builder, AI Engineering Infrastructure [Start date] - Present - Built DevBrain from scratch: a local-first memory, context, and orchestration layer for AI-assisted engineering across Claude Code, Codex, Gemini, and project-specific developer workflows. - Designed and implemented an MCP server with tools for project context, semantic search, session start/end lifecycle, breadcrumbs, graph traversal, and remote agent prompting. - Built a local Postgres/pgvector memory system using embeddings, background ingestion, summarization, cross-project fanout, graph relationships, and launchd-managed processing jobs. - Created an agent factory pipeline with planning, implementation, review, QA, fix-loop, and status-tracking phases, plus Agent Bus for authenticated remote agent execution. - Hardened the platform for reuse across machines and projects: externalized config, added doctor checks, wrote installation/architecture/docs, handled migration backfills, and drove test coverage to green during platform hardening. - Used DevBrain as a real internal platform, not a demo: it supports BrightBot development context, teammate onboarding, compliance-aware memory, cross-project recall, and long-running engineering coordination. ## Selected Systems And Architecture Work ### BrightBot Production Platform - Next.js UI, FastAPI API, PostgreSQL 16, pgvector, Alembic, Cloud Run, Cloud SQL, private VPC, Google Workspace integrations, Monday.com ingestion, HubSpot sync, Zoom workflow support, and AI agent tooling. - Enterprise patterns: RBAC, scoped APIs, audit-aware design, soft deletes, schema isolation, secure data migrations, rollback snapshots, and compliance review. ### DevBrain AI Engineering Platform - MCP tools, local embeddings, pgvector retrieval, graph memory, background summarization, cross-project fanout, launchd services, remote agent bus, and multi-agent engineering factory workflows. - Built to make AI-assisted engineering more reliable: persistent context, explicit session lifecycle, searchable decisions, issue memory, and reviewable technical rationale. ### Production Reliability And Data Integrity - Diagnosed migration drift between SQLAlchemy models and Alembic migrations, then shipped additive migrations and recommended parity tests to prevent recurrence. - Repaired production data defects with snapshot-backed, targeted updates and code-path fixes to prevent reintroduction. - Identified Alembic multi-head deployment failures caused by parallel branch work and established the merge-migration prevention pattern. ## Technical Skills Backend: Python, FastAPI, Pydantic, SQLAlchemy, Alembic, REST APIs, service-layer architecture, background jobs Data: PostgreSQL, pgvector, schema design, migrations, data pipelines, source reconciliation, embeddings, hybrid search Frontend: Next.js, React, TypeScript, Tailwind, dashboard and workflow UI, role-aware application surfaces Infrastructure: GCP Cloud Run, Cloud SQL, VPC, GitHub Actions, launchd, environment/config hardening, production debugging Security / Compliance: RBAC, OAuth, JWT, audit logging, HIPAA, FERPA, PHI minimization, soft deletes, access scoping AI Systems: MCP, agent tools, embeddings, retrieval, human-in-the-loop workflows, promptable internal tools, AI-assisted engineering pipelines Integrations: Google Workspace, Monday.com, HubSpot, Zoom, OpenAI/Anthropic-style LLM workflows ## Operating Style - I care about the interface: APIs, UI workflows, data contracts, and internal tools are product surfaces. - I move quickly, but I do not treat production data, auth, compliance, or billing logic casually. - I like ambiguous problems, especially when the real answer requires code, product sense, and operational judgment. - I build for where workflows are headed, then keep iterating as capabilities, user expectations, and failure modes evolve. - I use AI aggressively as leverage while staying accountable for architecture, correctness, safety, and taste. ## Education [Degree / program], [School] - [Year] [Optional certifications, coursework, or omitted if not useful] ## Additional Experience [Earlier role], [Company] - [Dates] - Add prior work here that supports backend systems, technical leadership, product ownership, operations, or customer-facing B2B work.