Rio de Janeiro · Forward-deployed engineer · Marketing for 15 years, building for the last 5
I'm Gabriel, aka Gabe. Forward-deployed engineer. Fifteen years working in marketing, and over the last five years, my work has turned to the building side, mostly applied AI. AI automation and agent plumbing are both my lanes.
I spot each team's gaps and bring systems to help them work better and faster. TL;DR & more deets here in this video.
| Project | Stack | What it does |
|---|---|---|
Tixmancer private · main focus |
 |
Event tickets bought by an agent end to end: HTTP 402 offer, signed QR ticket, one-time check-in, onchain receipt recovery on Base. 50 test files across unit, integration, contract, and browser e2e; Base Batches 004 application in. |
Maria private · milestone 1 |
|
The scam-checker Brazilian elders talk to on WhatsApp: forward anything suspicious, hear a voice verdict in PT-BR, family alerted when it matters. Building it for RevenueCat's Shipaton 2026. |
| Project | Lang | What it does |
|---|---|---|
| hedwig-sol ⭐ | |
Composable onchain roles for Solana. Grant roles, not keys. One canonical role store other Anchor programs share instead of rebuilding access control. 6 instructions, 21 LiteSVM integration tests, a documented threat model. Built on a Solana Foundation grant. |
| worldcup-pari-market ⭐ | |
Proof-settled World Cup prediction markets on Solana, using TxODDS validate_stat CPI and Merkle-anchored results. Live on devnet, 39 Rust tests. Demo. |
| kageb ⭐ | |
Private intent pooling for Solana: four real orders in, one aggregate trade out. |
| grimoire | |
Grimoire, the Solana Transaction Doctor. Paste a failed transaction signature and it decodes the failure (Anchor errors, custom program codes, compute budget, rent, PDA collisions) into a plain-English root cause and a concrete fix. Live demo + open source. |
| solana-ship-gate |  |
Auditor-grade pre-deploy safety gate for Solana programs: 4 deterministic checks, machine-readable PASS/FAIL, blocks unsafe mainnet deploys/upgrades via a PreToolUse hook. A Solana AI Kit skill. MIT. |
| patronus | |
Onchain-attested, runtime-scoped credentials for AI agents on Solana DeFi. |
| worldcup-settlement | |
An AI agent that bets on live World Cup matches on Solana, autonomously. |
| meteora-jup-sol-safety-monitor |  |
Read-only Meteora DLMM PnL and safety monitor with Telegram alerts. |
| solguard | |
Warns Solana users before they get rugged. Scores new pump.fun launches on deployer history, liquidity and mint authority, then posts the risk live. |
| superteam-academy | |
Open-source gamified learning for Solana devs, built for Superteam Brazil. Browser code editor, XP, leaderboards and onchain certificates. |
| Project | Lang | What it does |
|---|---|---|
| silvia | |
A confirmation-first food-ordering assistant for older adults, built on Base44. Chat instead of learning an app; 50+ users completed an order. |
| safeskill | |
One-click security audit for your MCP setup. One score, plain English, no CLI knowledge required. |
| wingman | |
AI Expert Marketplace. Install experts (Design, Code, Marketing) into Claude Code. |
| clawmanship | |
Reviewable decision before a community skill enters a local bundle. |
| agenthub | |
Monad-native AI agent orchestration platform. |
| hermes-aria-theme |  |
Minimalist black and white dashboard theme for the Hermes agent. Green for success, red for errors, nothing else. Greek-inspired, with the Omega on every session card. |
| Project | Lang | What it does |
|---|---|---|
| walletbrief-monad | |
Persistent Monad wallet briefs with human-approved, revoke-only EIP-7702 execution. |
| yieldpilot | |
Your AI copilot for cross-chain yield optimization. Built for the Chainlink Convergence Hackathon 2026. |
| Project | Lang | What it does |
|---|---|---|
| beat-claude-engineer-004-analytics-pipeline | |
Real-time analytics pipeline design for the Beat Claude engineer-004 challenge: fixture anomaly detector, test suite, hot-path benchmark, and doc/evidence consistency checks. |
| ecommerce-sales-analysis | |
E-commerce sales analysis with Python (pandas, matplotlib) and SQL. RFM segmentation, revenue trends, product performance. |
| solana-narrative-tracker | |
Spots emerging Solana narratives early. Reads 20+ KOLs, 20+ repos and 25+ protocol TVLs, clusters the signals, and ships build ideas every fortnight. |
| defillama-tvl-anomaly-detector | |
Scans DeFi protocols for TVL that looks wrong. Zeroed, dropped 50% in a day, spiked 300%, or flat for 60 days. Catches broken adapters, exploits and depegs. |
Also: atelier-gabriel-paz. A landing page I built for my friend Gabriel Paz, a painter in Botafogo, Rio. He had no website, so I made him one. Hand-written HTML, no build step.
Off GitHub: a six-agent system runs my daily marketing operations, research, drafts, review, planning, with a human sign-off at every step. 27 of 30 content cycles held above my quality threshold, and the daily brief hit 97% accuracy across 31 traced runs.
A high-severity find in Anchor's LazyAccount code, v0.31.0 to v0.32.1: a silent data-corruption bug, disclosed through GHSA-wqrf-hpg4-vgv8, fix published as a public PR, bounty paid. Also a winning Superteam submission, the Solana Vault Standard Extension: 4,398 lines of code across 34 instructions.
Note
Additional findings are under review across active Immunefi programs. I don't publish details of an open or rejected report. Program terms exist for a reason.
- 91.9% of active allocators (260 wallets) migrated V1 to V2 at Octant, on zero paid spend.
- 13K followers and 9M organic impressions in 9 months on @octantapp.
- Onboarding conversion moved from 30% to 70% at noder., across 100+ experiments.
I write about proof, evals, and AI-assisted marketing at gabeonchain.com.
- Marketing has a proof problem
- How I stopped winging it on evals
- Keep the judgment, cut the slop
- The fence is the job
gabeonchain.com · X @gabe_onchain · LinkedIn