Skip to content
View ethan-haas's full-sized avatar

Block or report ethan-haas

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.

Content in all repositories owned by your account will be closed.
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
ethan-haas/README.md

Ethan Haas

Applied AI Engineer · I build production LLM and agent systems, and I care most about the hard half: making the output verifiable, not just plausible. The model proposes. Tested code decides.

🔗 ethanhaas.dev · LinkedIn · ethanzhaas@gmail.com · Open to contract or full-time work


What I build

  • Agentic systems · multi-agent orchestration (LangGraph, custom harnesses, MCP, tool use) built behind verification so improvements are earned, not hallucinated.
  • RAG & retrieval · grounded answers with page-level citations and safe-refusal patterns, so every claim can be checked rather than trusted.
  • LLM evaluation & guardrails · golden sets from expert-handled cases, accuracy measured not assumed, low-confidence output flagged to a human instead of shipped.
  • Full-stack AI products · production B2B SaaS end to end (auth, multi-tenant data, billing, audit trails) and document intelligence where a wrong number has consequences.

Featured work

self-improving-agent-harness · an autonomous multi-agent system that improves its own code behind a 7-tier verification gauntlet, including a fresh-context critic and a sealed holdout. Inner research loop ported from Udit Goenka's autoresearch (MIT, based on Karpathy's work); the meta-improvement architecture and verification stack are my own.

🟢 OBBBA Tracker · a live, paid B2B SaaS I build and operate. Tip & overtime tax-compliance for tipped-industry employers: automatic TTOC codes, FLSA overtime, W-2 Box 14 exports (ADP/Gusto/QuickBooks), multi-tenant access, billing.

financial-statement-extractor · verifiable document AI: government audit PDF to validated JSON, re-deriving every total (foot / crossfoot / articulate) with a golden-file test. 25 checks, 0 exceptions on a published report, and it catches an injected wrong figure.

grantledger · B2B SaaS that auto-categorizes nonprofit grant spending into 2 CFR 200 budget categories and generates audit-ready compliance reports. Next.js · Supabase · Stripe · OpenAI.

pacman-cpp · native C++20 + SDL2 Pac-Man built from scratch: co-op multiplayer, four ghost AIs, procedural audio, cross-compiled to an ARM handheld, with sanitizer/coverage/strict build presets and tests.

Stack

Python · LLM APIs (Claude / Gemini / OpenAI) · AI agents / LangGraph / MCP · RAG / vector search · LLM evaluation & guardrails · FastAPI · TypeScript / React / Next.js · C#/.NET · Docker · pdfplumber / PyMuPDF / OCR

Background

B.B.A., Finance · University of Cincinnati (Cum Laude) · CFI FMVA & BIDA. I came up through finance and taught myself to build, which is why I come at every model output asking whether it's actually right, exactly what regulated, high-stakes work demands.


Open to Applied AI Engineer, Forward Deployed Engineer, and AI Software Engineer roles, remote · say hello.

Pinned Loading

  1. self-improving-agent-harness self-improving-agent-harness Public

    Autonomous multi-agent system that improves its own code behind a 7-tier verification gauntlet (sealed holdout + fresh-context critic) — improvements earned, not hallucinated.

    PowerShell

  2. financial-statement-extractor financial-statement-extractor Public

    Government audit PDFs → validated structured JSON. Deterministic extraction + foot/crossfoot/articulate tie-out + golden-file test. Tested code does the math, not the model.

    Python

  3. neurofour neurofour Public

    Connect 4 strength-per-byte arena — play 20 agents, exact-solver analysis, and a strength-vs-cost Pareto leaderboard where a 0-byte search agent is champion. FastAPI + React.

    Python

  4. grantledger grantledger Public

    B2B SaaS automating federal grant compliance for nonprofits — AI expense categorization into 2 CFR 200, budget-to-actual tracking, audit-ready reports.

    TypeScript