Skip to content
View evanjcosgrove's full-sized avatar
⚙️
⚙️

Block or report evanjcosgrove

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
evanjcosgrove/README.md

Evan Cosgrove

Forward-Deployed Engineer · NYC — I embed where the problem is, ship the production code that solves it, and prove it works with evals.

Brooklyn, NY · evanjcosgrove@gmail.com

Shipped agents into production? Mindara, the AI event-planning platform I co-founded with Justin Casso: 1,700+ commits, 250+ DAU, Claude API in production, Django + Expo/React Native on GCP/Terraform.

Post-sales implementation? Google: $25M+ incremental growth on a $90M+ Ads portfolio, working hands-on with agency and client engineering teams on API integration and data-flow debugging.

The pattern I can't turn off: when a tool I depend on falls short, I build the better version. A 1M-token context regression in Claude Desktop sat open for a month, so I diagnosed and patched it. Every AI coding surface kept forgetting what the others knew, so I built cross-tool memory. My Toast territory needed prospect intelligence that didn't exist, so I built it as a Retail AE — not an engineer.

What I ship

Repo What it proves
claude-cowork-1m-patch Diagnosed and resolved a month-old open community regression in 48h: Electron .asar byte-patching, integrity-chain rebuild, macOS re-signing
fde-case-studies The production work that lives in private repos: Mindara, a legacy-to-Symfony-7 CRM modernization (consulting engagement), the Toast retail prospecting copilot, cross-tool agent memory

Currently building

mirrorloop — a self-validating dev team for Claude Code: frontend agents check their work against a live iOS simulator, backend agents check the APIs, and an eval harness scores the whole loop.

day-one — an enterprise onboarding copilot: an agent team that reads an API's docs and sandbox, then produces a working, tested integration with a correctness report.

Both land here when they meet their definitions of done.

Stack

Python (Django, FastAPI) · TypeScript (React, React Native, Bun) · LLM systems (Claude/OpenAI/Gemini APIs, MCP, agent teams, eval harnesses, pgvector/ChromaDB) · GCP + Terraform + Docker + GitHub Actions

Before engineering full-time

Nine years of enterprise experience: Google (Ads API work with agency and client engineering teams), Toast (retail POS, Brooklyn), Remesh ($945K+ enterprise SaaS ARR with McKinsey, BCG, and Accenture), and Jet.com/Walmart eCommerce ($32M P&L). I speak fluent stakeholder.

Pinned Loading

  1. claude-cowork-1m-patch claude-cowork-1m-patch Public

    Restore 1M context window in Claude Desktop's Cowork (macOS)

    Shell 2

  2. fde-case-studies fde-case-studies Public

    Four production case studies: agents shipped, systems modernized