June 2025 – Present
Homebot.ai — Senior Software Engineer
Remote, Denver, CO
Lead, Customer Experience team (6 engineers, 4+ applications); core member, AI Intelligence team; reporting to the CTO and CPO. First engineering hire of the incoming CTO, and sole engineer and product manager on the AI platform and the market-intelligence rewrite.
- Cut support tickets 70-80% and reduced response times from months to hours by shipping proactive and reactive AI agents across support, customer success, and engineering.
- Replaced a product that had taken a team of 4-8 engineers roughly 2 years to build, shipping the full rewrite in 6 months as sole engineer and product manager, and cutting infrastructure and vendor spend $5,000+/month.
- Partnered directly with 3 enterprise design partners across lending, real estate, title, and insurance - Guild Mortgage, Newrez, PrimeLending, Huntington National Bank, KeyBank, and Windermere among them.
- Opened title and insurance as new segments through customer discovery and feature development, unlocking a serviceable market of ~$10.2M MRR / $122M ARR, and partnered with the ALTA title registry.
- Built the AI agent backend behind the company's customer chat and support products (Mastra, AWS Bedrock), including a federated GraphQL gateway across 12+ data subgraphs with per-tenant isolation at the schema level.
- Delivered the first federated-GraphQL self-query tool so the AI agent could answer live customer questions against production data, then led its evolution to SQL self-query with a persistence layer for auto-updating customer dashboards.
- Created the evaluation harness for both agents - scenario suites, scoring rubrics, and regression gates - so prompt, model, and tool changes ship on measured output quality, not opinion.
- Ran the agents across AWS Bedrock, Anthropic Claude, and OpenAI with per-task routing and fallback, tuning retrieval over pgvector and Elasticsearch, tool-result trimming, and conversation compaction to hold quality and cost inside the context window.
- Cut average tokens per model call 37% by moving workflows from tools to skills, with prompt caching and reusable generative-AI component libraries; enforced guardrails, PII redaction, and explicit database grants over what each agent can reach.
- Automated engineering workflows across 20+ repos with Claude Code and Codex - custom AI skills, plugins, and agentic workflows for code review, runbook execution, documentation, and day-to-day Kubernetes and GCP operations.
- Corrected 9.5 million misattributed records and removed a query-cost blowup from a customer-facing lookup path, redesigning the data model on a platform of 131TB active Snowflake databases and a 2.3TB Postgres store.
- Tripled the PR merge rate from 2024 to 2026 by reorganizing the team into small cross-functional pods; reviewed 480+ pull requests and authored 9 design documents and 7 runbooks that became the org's source of truth.