Richard Boneff-Peng

• Built a pre-implementation interview methodology for autonomous coding agents. Before any code is generated, the system interviews the engineer, pre-answering every question that would cause the agent to stall or make assumptions. The engineer walks away during implementation and comes back to a first-pass PR that's close to mergeable. Ships as a Claude Code plugin with six skills, a compliance subagent, and a telemetry server. Every skill is developed through eval-driven development — 90 evals with 232 assertions. No improvement ships without eval coverage.
• 15 engineers across 7 teams, 92 interview runs over two months of telemetry. 31 PRs from 30 distinct tickets. An EM started using the interview questions to triage tickets — one ticket assumed to be development-ready got kicked back to discovery based on the gaps the system surfaced. A senior engineer now uses it on every ticket in a multi-ticket project.
• Built a telemetry pipeline that layers tool-specific measurement on top of existing org-level DX metrics. Generic schema — any tool can emit into the same system without touching developer machines. Ran the first comparison of interview-methodology PRs against the repo baselines across 5 repos. Methodology PRs are 1.9-5.4x larger (greenfield feature work, not small fixes) with proportionally better first-draft quality: zero median revisions on one team, comparable per-line revision rate on another despite 5.4x the code volume.
• Built a convention-over-configuration platform where engineers define their service in one YAML file and get production monitoring, CI workflows, Kubernetes manifests, and catalog metadata generated from it. Org-level policy changes apply once instead of per-service. 5 services in production, 12 in progress. First deployment surfaced a memory leak within two minutes through an auto-generated saturation alert.
• Co-authored the org's AI onboarding guide at the start of the adoption rollout — the entry point engineers used to get started with AI-assisted development in a way that fit the org's workflows. Wrote a companion guide for building testable Claude Code skills through eval-driven development. Designed the technical interview scenario bank used to hire two Staff Engineers.

• Ephemeral environments were the #1 requested feature. I scoped and built the platform in 6 weeks. Engineers tag a PR to spin up isolated services against a shared integration fleet, test without claiming shared staging time, and debug directly in pods. Once the POC was running, the team expanded service coverage. Adoption was immediate once we hit critical service coverage.
• This replaced 3 snowflake staging environments that blocked QA and caused deploy conflicts. Ephemeral sandboxes on Kubernetes rebuild automatically on every push. CI/CD pipelines use a convention-over-configuration architecture so teams could adopt incrementally.
• Introduced schema migration safety practices where none existed: rollback on every migration, structured logging, and smoke tests running against mirrored production data before deploying.

Genomenon Mastermind

• Designed and operated a 16-node Elasticsearch cluster indexing ~16 million articles. Geneticists depended on this data for rare disease diagnosis. This was early Elasticsearch — before Kibana, before management interfaces. I built the ingestion pipeline, index migration tooling, and monitoring myself. The pipeline handled both incremental weekly updates and full corpus reindexing when performance needs or data structures changed. Dealt with actual data loss from platform bugs and recovered.
• The Ruby on Rails application couldn't parse the large JSON payloads from Elasticsearch fast enough. I built a Go proxy that handled enrichment and auth in front of the cluster — orders of magnitude improvement in reliability and response time without rewriting the Rails application.
• Geneticists were manually searching PubMed, choosing articles from abstracts, each person carrying their own reading bias. I worked with the Chief Science Officer to build a search and curation interface that let them triage articles by genetic sequence across the full 16-million-article corpus. What took months of curation took days. The capability changed the company's product direction.

Consulting Engagements

• Served as acting CTO for multiple startup clients — architecture decisions, product scope, investor timelines, and founder expectations all landed on me. Most projects I was solo or paired with a junior engineer. Mentored junior founding engineers on testing, documentation, and code quality while shipping under pressure.