I fix the CRM your team outgrew.

Point it at a messy CRM and it hands back one you can trust. Seven phases run in order: ingest, dedup, normalize, enrich, apply your rules, pause for a human to approve the risky calls, then write back. It runs a local model for the cheap work and escalates to Claude or GPT-4o only when a decision is hard, so you are not paying frontier prices to lowercase an email. This is the flagship, and it is still being built.

The context layer inside Harness. Fragmented sources kept current and wired to executable skills, so the pipeline reasons from an up-to-date picture instead of relearning everything each run. This is the bet: agents take the operational grind, people move up to designing the system.

Where the data discipline meets strategy. I took a B2B company from a blank CRM to a full revenue engine: market sizing and segmentation, lifecycle stages with real entry rules, a lead-source taxonomy that survives reporting, deal stages named for buyer outcomes with inspectable exit criteria, and the marketing-to-sales-to-success handoffs where pipeline usually leaks. Documented as a build handbook and a one-page revenue-engine blueprint, not a slide about one.

Enriched Clay records land in HubSpot clean. It matches identity before it writes, resolves conflicts instead of overwriting good data, and never doubles a record it has seen before. It is deployed and running, not a screenshot of a dashboard.

A client had the same company showing up three ways across job boards and no way to tell one real lead from a duplicate. I built the pipeline that fixes it. n8n scrapes the sources, Clay normalizes the names so "Avery Dennison Corporation" and "Avery Dennison" become one company, fuzzy matching kills the duplicates across roughly 20,000 contacts, then enrichment and outreach take over. Shipped, with recorded walkthroughs.

Proof the skill travels outside SaaS. I took a construction firm's legacy accounting system and rebuilt it from scratch into a working web app with Claude Code. Messy real-world records, hard consistency rules, an industry that never gets this treatment. Same instincts as the CRM work, different vertical.