I build the checks that catch what is breaking quietly.

I own the internal systems layer: the workflow logic, the integrations between it and everything else, and the reporting that has to make the result legible to people who were not in the room. Roughly sixty active project directories, two-thirds of them touched in the last month — a maintained estate rather than a pile of scripts.

• Scope and deploy technical workflows across customer success, onboarding and internal operations using SQL, CRM APIs and multi-department tooling.
• Built and maintain SQL-backed reporting for subscription logic and billing reconciliation.
• Lead employee onboarding and offboarding workflows tied to access control and system provisioning.
• Design and implement CRM automations across customer success, sales and partnerships.
• Support internal infrastructure through custom workflow development and scripting — ingestion services and scheduled jobs on managed cloud infrastructure, warehouse-backed reporting behind access control, and operational microservices. Python-dominant, TypeScript for user-facing work, around 140 test files across the estate.
• Built an LLM-assisted compliance service: it transcribes calls and scores them against a rubric, deterministic checks first and human review after, with the model never given the last word.

Owned cross-department process infrastructure and automation across customer success, product, sales and operations — the reporting, tooling and workflow design that made churn something the company could see before it happened.

• Cut client churn from 30% to under 10% through process and engagement redesign backed by real-time risk scoring.
• Built and ran SQL-driven credit, billing and subscription reporting.
• Coordinated ticket workflows across success and transaction-coordination operations, halving issue resolution time.
• Served as the customer-success stakeholder for CRM, internal tooling and platform workflows.
• Wrote and deployed the customer-success playbooks and onboarding documentation.
• Kept every system running through a 30% headcount reduction. Nothing broke.

• RunEcho — checks whether a coding agent's claims about code are actually true instead of trusting them. Used daily for five months.
• terse — compresses AI tool output losslessly: 58% fewer tokens across 365,144 tokens of real API payloads, measured against the TOON format, which regresses to −7% on the same corpus. Public, MIT.
• Lodestone — local retrieval over years of working notes, and the evaluation that showed half of it did not work. Published the null result.
• Frostline — spatial models over 50,083 well records, and a business built around what they can honestly predict.
• Reported 32 bugs in a code-analysis tool I use daily, including two race conditions reproduced on demand with purpose-built harnesses.
• Two servers other tools call: one issues credentials without exposing the value to the caller, one stores decisions and refuses anything unreviewed.
• Three applications deployed end to end — a task scheduler with calendar sync, an e-signature and payment intake tool, and a lead-intake product for a client.

• Handled customer service for 70+ partners, 3,000+ partner clients and 400,000+ end customers, responding to every ticket within two hours.
• Tested issues for reproducibility before escalating, documented what would not reproduce, and ran the weekly session that fed that back to product.
• Wrote the FAQ and knowledge-base documentation, and carried customer feedback into product recommendations.

• Ran day-to-day operations under five service contracts, handling up to 70 calls and 100 emails a day.
• Created and documented the processes underneath — refunds, chargebacks, phone sales, fulfilment.

• Managed complex caseloads for four attorneys.
• Worked across departments to keep discovery moving, so attorneys had time to prepare for hearings.