// 00 · WHOAMI · REMOTE · EMEA

OMAR
GARCIA.

Senior Software Engineer & Tech Lead.

Hire me to build your AI agent →View work

// streaming response from claude-opus-4.7·github.com/ogarciarevett

// 01 SELECTED_WORK

Selected work.

Three or four engagements that explain what I actually do, not a job-history dump.

// 01_ HYBRID L1 DEX (NDA)2025

Hybrid DEX & Matching Engine

WHAT: Tech Lead on a hybrid on-chain/off-chain DEX built on a high-throughput L1, including a low-latency matching engine and MCP servers for trading automation.
HOW: Rust core, deterministic order book with parity replay against the chain. Pre-confirmation path tuned around the chain's optimistic block production. MCP tool surface exposes order placement, market stats, and risk to Claude-driven agents.
IMPACT: ~300 μs pre-confirmation, ~3 ms full SPOT round-trip, 12.5k TPS sustained on internal benchmarks.

RustGoTypeScriptMCPSolidity

// 02_ GAME7 DAO · SUMMON2023 — 2025

Multitenant Governance AI Platform

WHAT: Founder Engineer of Summon, a multitenant DAO platform with an AI governance copilot and the Summon MCP server.
HOW: RAG pipeline against on-chain proposals + forum threads, exposed to non-technical admins. Firecrawl-driven AI theming generates branded surfaces per tenant. Cross-VM rewards contracts target EVM, Solana, and Move from one source of truth.
IMPACT: 100+ smart contracts deployed across three VMs. The Summon MCP server is one of the earliest production MCP integrations for DAO governance.

TypeScriptSolidityMCPRAGMoveAnchor

→ G7DAO/summon-api-suite

// 03_ LEMONCASH2021 — 2023

Crypto Exchange Backend (founding eng)

WHAT: Founder Engineer of the LemonCash exchange backend — order routing, custody, fiat rails, KYC/AML.
HOW: Node + TypeScript services, Postgres + Redis, idempotent withdrawal pipeline, internal HSM integration. Designed the matching layer that survived a 40× user spike during launch.
IMPACT: Backbone for one of LATAM's largest crypto-fiat onramps; team of 4 → 30 within the first year.

TypeScriptNodePostgresRedisAWS

// 04_ FORTE GROUP · BLOCKFI2019 — 2021

Principal-level systems work

WHAT: Principal Engineer at Forte, embedded with BlockFi on consolidation of its yield-generating product surface.
HOW: Service decomposition, Kafka event spine, internal SDK for asset accounting. Owned the tooling for the audit-track migrations leading into the 2021 product expansion.
IMPACT: Halved the turnaround time for new asset listings; cut on-call paging volume on legacy services by ~60%.

GoKafkaAWSPostgresPython

// 02 OPEN_SOURCE · STANDARDS

Open source & standards.

Co-author of two ERC standards. Smart contract auditor at Code4rena since 2023. Contributor across Game7 and zkSync.

// 365_DAYS · ogarciarevett · 1,710 contributions

lessmore

longest streak: 25d · current: 3d

// 02.1 PINNED_REPOS

ogarciarevett / mcp-trading-tools

MCP servers exposing structured tools to Claude — orderbook, risk, fills, retrieval.

TypeScript★ 62

G7DAO / summon-api-suite

MCP server + governance API — one of the earliest DAO MCP surfaces.

TypeScript★ 84

ethereum / EIPs

Co-author of ERC-7537 (Soulbound Tokens for DAOs and Web3 Games).

Solidity★ 21k

matter-labs / zksync-web-era-docs

Contributions to zkSync developer docs and tutorials.

MDX★ 1.1k

moonstream-to / web3

Open-source Web3 indexer and analytics primitives.

Python★ 230

// 02.2 STANDARDS · RESEARCH

ERC

7537

● DRAFT

co-author

Soulbound Tokens for DAOs and Web3 Games

Non-transferable tokens for governance roles, contributor reputation, and game-state ownership. Defines a minimal standard with revocation, expiration, and on-chain attestations.

→ Read on ethereum-magicians.org

PAPER

2026

● PUBLISHED

co-author

Block Update Digests: Membership Proofs Without a Global State Tree

Authored with Cody Littley and Alejandro Ranchal-Pedrosa. BUDs are an incremental authenticated data structure that scales membership proofs with per-block write volume rather than total state size — no global state trie on the critical path.

→ Read on ethresear.ch

// 02.3 SECURITY · AUDITS

Smart Contract Auditor · Code4rena · since Jun 2023

Findings across 9 contests. Specialty: cross-VM bridge invariants and matching-engine economic safety.

View profile →

// 03 RESEARCH · WRITING

Research & writing.

// PhD · MACHINE_LEARNING2026 — PRESENT

University of Cambridge

PhD in Machine Learning Engineering. Focus: tool-use efficiency in agentic LLMs and ZK-provable inference for on-chain agents.

// MSc · ARTIFICIAL_INTELLIGENCE2024 — 2026

University of Zurich

MSc in AI. Thesis on retrieval-augmented governance copilots (the prior art for the Summon RAG pipeline).

// WRITINGCOMING SOON

Notes — coming soon.

Drafts in flight: MCP server design patterns, matching-engine determinism on high-throughput L1s, and ZK + ML for verifiable inference.

// 04 SERVICES · HIRE_ME

Hire me to set up your personal AI agent.

I run two-week, fixed-fee engagements to set up production-grade Claude + MCP agents: tool surface design, RAG plumbing, observability, and on-call runbook. Most teams ship a working internal agent by week two.

If you've got a knowledge surface (docs, ticketing, repos, on-chain data) and you want Claude to actually do work against it, that's the engagement.

omar@mcp — claude-haiku-4.5 — 80×24● connected

// click a command above to run a sample tool call

✓ session established · 14 tools registered

LinkedIn DM →

// 05 ABOUT

About.

I work remotely across EMEA, currently doing the PhD at Cambridge. Publicly I started with open-source contributions to React at Meta, but my actual first stack was Angular 1, MongoDB, Postgres and Node.js 0.4.X back at my second job — full-stack from day one. From there: backend systems, crypto exchanges, smart contracts, standards work, and over the last two years deep into agent infrastructure and LLM systems.

The connecting thread is the same as it's always been: I like systems where the boring details — determinism, latency, replay — decide whether the cool feature actually works in production.

“Most "AI agents" fail in production because they were prototyped against optimistic tools. The MCP layer is where you decide if your agent is an assistant or an outage.”

// YEARS_SHIPPING: 13+
// PUBLICATIONS: 2
// CONTRACTS_DEPLOYED: 100+
// MATCHING_LATENCY_μS: ~300