Edward Federbush

AI Product Engineer·Three Startups, Two Acquisitions

I've been building software products since 2001, from first engineering hire under the founding CTO at GLG through co-founding two acquired startups. Today I build AI into the center of a product: an assistant that runs a real business from first call to paid invoice, with every model chosen by measurement and evals behind every change. The through-line: I build directly with the people using the product — from teams inside Fortune 500 companies to small field-service operators.

Current Focus

Wipjet, an AI-native platform for home-service businesses — and the engineering that makes an assistant trustworthy enough to run one: tools designed for models, head-to-head model bakeoffs, behavior and prompt-injection evals, and a human confirming anything that touches money or a customer.

Experience

Independent

AI Product Engineer · 2022 – Present

Independent product engineering since 2022; since 2025, focused on building Wipjet.

AI-Native Field-Service Platform, Wipjet · 2025 – Present

Founding engineer on a multi-tenant platform that runs a home-service business from first call to paid invoice, with one AI assistant that follows the work the whole way. The founding-pilot phase is complete, and the company is moving into go-to-market.

  • One assistant across the lifecycle: voice agents and web chat take the inquiry, and the assistant accepts the lead, drafts and sends the estimate from the shop's own rates, answers questions over live schedule, parts, and invoice data, and drafts the invoice — a human confirming anything that touches money or a customer.
  • Tools designed for a model — an intent-level API with idempotent writes — exposed in the app and through a remote MCP server with OAuth, so an owner can run the business from Claude; tenant isolation by Postgres row-level security under every tool call.
  • Each model chosen per task from head-to-head bakeoffs on quality, latency, and cost through one AI gateway, with swaps reverted when production disagrees with the golden set.
  • Evals gating every prompt and model change: behavior evals in which a model drives the real tools against a seeded shop, graded by code checks and calibrated judge models, and a 59-case prompt-injection suite across chat, assistant, and MCP.
  • Every call traced through OpenTelemetry into Langfuse and attributed per business — cost, latency, and prompt-cache hit rates by tenant and surface.
  • Built through a spec-driven agent practice: verification designed around how coding agents actually fail, fed by a ledger of real failures that every new spec draws from.
  • Stack: React 19, Next.js 16, TypeScript, Postgres, Drizzle ORM, Vercel AI SDK, Model Context Protocol, Langfuse, Stripe, Vercel.

Capital One

Discovery & Prototyping Engagement

Product Experience Consultant · 2021–2022

  • Brought in by the head of a new unit to research and design its first product, a co-brand credit product for direct-to-consumer e-commerce merchants; the engagement grew into designing an enterprise-wide platform to host it.
  • More than 40 hours of interviews with merchants and consumers, competitive research, and rounds of prototypes, through handoff to internal engineering in a large regulated enterprise.

Docurated

Acquired by Quark Software (2018)

Co-Founder & Full-Stack Product Engineer · 2012–2018

  • Co-founded SaaS platform for enterprise sales enablement (TechCrunch Disrupt finalist); led product experience and engineering from inception through acquisition, building across desktop, mobile, and embedded Salesforce apps.
  • Partnered directly with Fortune 500 clients (Netflix, J.P. Morgan, IBM) on custom implementations and white-labeled workflows.
  • In parallel (c. 2011 – 2013), built and ran a secure screening platform used by 6,000+ Academy members for Oscar voting and 100,000+ SAG members for SAG Awards voting, plus studio screenings at Sundance and Cannes; later wound down.

After the acquisition, took a planned break (2018–2021) following two back-to-back startups.

Boliven

Acquired by Cambridge IP (2011)

Founding Member & Full-Stack Product Engineer · 2008–2011

Employee #4 at this GLG-alumni-founded startup. Full-stack Rails role from early product through acquisition.

Gerson Lehrman Group (GLG)

Enterprise Expert Network

Engineer → Director of Product Design · 2001–2008

  • Founding CTO's first engineering hire. Early work was full-stack, building the early product as half of a two-person team.
  • Transitioned to product leadership as the company grew, building and leading a ten-person cross-functional team of designers, UX engineers, and researchers.

Education

Boston University — BA, Philosophy