Rasul Shaikh

Turns a messy 71-row CRM export into 59 accounts, removes 11 customers, Do Not Contact and staff before any paid step (0 credits), tiers the rest on a dated pricing visit plus fit, and lands 3 validated leads once each in a paused table. 94.5 credits in total. Tradeoff: I enriched every candidate early, so the next fix is a free fit check before enrichment.

Role map, research, signal, gap, angle, then a draft under 90 words, and a second model checks every draft for facts and spam risk. Built at Omnibound AI. 1.5% or more positive replies across 150+ domains. Tradeoff: an extra model call per draft, in exchange for no unsupported claims.

Every recorded outside call becomes one record tied to the right client, with a summary a person reads first. 22 n8n workflows, 15 live, 1,159 tests, 31 recordings merged into 15 meetings. Atlas, the task agent on top, answers create, update or skip for each action item: first test on 3 calls, 30 items, nothing written. Tradeoff: tasks and CRM write-back stay off until a person approves.

Inbound qualification went from 4 hours to under 3 minutes, with a branded brief on the AE's desk before the first call. 8 n8n workflows, a 5-provider enrichment waterfall and a B2B gate, 698 unit and 143 end-to-end tests. Tradeoff: five providers cost more per lead than one source, in exchange for fewer missed contacts.

Claude Haiku classifies every LinkedIn and email reply and drafts an answer. A person decides, and nothing sends on its own. Multi-tenant Supabase with row-level security, 183 commits in 16 days. Tradeoff: every reply waits for a human decision, which is slower and safer.