MS Builder Packet
Full-stack products, AI systems, real-time infrastructure

Mathew Stevens builds production software end to end.

Full-stack engineer shipping live products across web, mobile, backend systems, AI-enabled workflows, SDKs, and developer tools. I build real-time infrastructure, security-sensitive workflows, cross-platform apps, and polished user-facing product experiences.

Resume PDF GitHub LinkedIn X / Twitter

Live products

Live products with real users.

Two shipped production products: one for trading execution and risk, one for user-approved AI-agent signing across real wallet surfaces.

Agentic

First multi-wallet AI-agent signer across web, mobile, CLI, and desktop
Non-Custodial AI Agent Review proof

First multi-wallet AI-agent signer, live at agentic-signer.com with Apple App Store, Google Play, verified app-store surfaces, web app, CLI, and desktop releases. Agents propose chat-driven wallet actions, plans, and external payment requests; users approve or deny with their real wallet; keys never leave the wallet. One WalletBackend powers the chat/app flow, MCP tools, Vercel AI tools, A2A AgentCard metadata, AP2/ACP/MPP payment-adapter flows, CLI, desktop, and web surfaces.

  • First multi-wallet AI-agent signer with Apple App Store, Google Play, web, CLI, and desktop release surfaces.
  • 1250+ users, 40 reviews, and a 4.2-star rating across shipped product surfaces.
  • MCP/Vercel AI tools, A2A AgentCard metadata, AP2/ACP/MPP payment-adapter flows, approval polling, transaction simulation, and sign-and-send.
Website App Store Google Play dApp Store

SolPulse

Real-time trading platform
Live trading system Review proof

Solo-built production trading platform with multi-mode scanning, DCA, limit orders, smart TP/SL exits, 35 intelligence pillars, multi-DEX routing, priority fee controls, and institutional risk controls exposed through a consumer UI.

  • 1000+ users, 26 reviews, 4.3-star rating, Apple App Store, Solana dApp Store, and web releases.
  • Execution path documented for 1,200 concurrent transactions and 10ms monitoring intervals.
  • Seed Vault and Mobile Wallet Adapter signing flows keep keys inside user wallets/devices.
Website App Store Google Play dApp Store

Infrastructure repos

AI systems, SDKs, reliability tooling, and product infrastructure.

Public repositories showing transferable systems work across AI, reliability/evaluation tooling, SDK design, native/mobile bridges, security tooling, and game-engine integrations.

Agent Eval Foundry

Benchmark production system with 25 task packages. Nine selected finalists: seven at 6/6 and two at 5/6 agent failures (reward=0), each across Codex and Claude trials.

TypeScript / Docker / HarborGitHub

Capability Record Replay

Deterministic automation runtime where an LLM discovers a legacy back-office workflow once, then typed capabilities replay model-free through browser and terminal surfaces.

TypeScript / Playwright / CLIGitHub

Klavis TB3 Durable Outbox

Terminal-Bench 3 eval where 6/6 Claude Opus and OpenAI Codex trials failed clean reward-0 on UNKNOWN outcomes, receipts, revocation, exactly-once execution, and audit legality.

Python / DockerGitHub

ConfigPilot

AI Fund (Andrew Ng) build challenge, one of 12 engineers invited: B2B decision engine gating LLM recommendations behind deterministic margin, inventory, and policy checks.

Python / ReactGitHub

Crashpoint

Crash/recovery conformance kit for agent runtimes, measuring exactly-once external side effects through an out-of-process ledger across persistence barriers.

Python / Temporal / DBOSGitHub

Durable Agent Outbox

TypeScript infrastructure for AI agents that call real external systems without duplicating payments, transfers, tool calls, or other side effects after crashes and retries.

TypeScript / SQLite / PostgresGitHub

Cheat-Oracle

Defensive verifier-hardening audit showing how in-container answer-key detectors undercount benchmark reward hacking, plus a tamper-evident out-of-container detector.

Python / Docker / Eval IntegrityGitHub

Judge-Artifact

Eval-integrity audit measuring how shipped success judges over- and under-credit identical agent transcripts, including prompt-injection ASR inflation on AgentDojo data.

Python / Inspect / AgentDojoGitHub

Agent Context Containment

Prompt-injection containment engine that tracks value provenance and blocks untrusted web, email, document, or retrieval content from steering dangerous tool arguments.

TypeScript / Policy EngineGitHub

Model Regression Sentinel

CLI drift detector for pinned model aliases: reruns frozen prompts, compares new outputs against baselines, and confirms regressions before failing CI.

TypeScript / CLI / EvalGitHub

Toolcall Risk Classifier

Pure-Python pre-execution risk scorer and eval harness for proposed agent tool calls, with leakage audits, conformance tests, and cross-source transfer analysis.

Python / Safety EvalGitHub

TxShield / Solana Tx Guard

Open embeddable Solana transaction-safety layer with pre-sign drain detection, static rules, optional simulation, and atomic guard.

TypeScript SDKGitHub

Unreal Solana MWA

Production-ready Unreal Engine 5 plugin wrapping Solana Mobile Kotlin clientlib through C++ JNI and Blueprint APIs.

UE5 pluginGitHub

Cocos MWA SDK

First Solana Mobile Wallet Adapter SDK for Cocos Creator, with native Android bridge, encrypted association, and demo game.

Cocos CreatorGitHub

iOS Wallet Adapter

Swift/CryptoKit wallet adapter for Phantom, Solflare, and Backpack deeplink signing with encrypted URL callback flows.

Swift packageGitHub

Capacitor Solana MWA

Capacitor bridge for Solana Mobile Wallet Adapter, giving hybrid apps native Android wallet support.

CapacitorGitHub

Godot MWA Example

Grant-backed Godot example app with multi-wallet flows, reconnect coverage, SIWS, and Seeker hardware verification.

Grant workGitHub

Unity MWA Example

Grant-backed Unity example app that surfaced SDK issues under real Seeker hardware testing and mainnet-beta flows.

Grant workGitHub

Upstream proof

PRs accepted by SDK maintainers.

Merged PRs are written by delivered outcome instead of raw method names, so the section reflects the actual engineering scope.

What I work with

TypeScript, JavaScript, React, Node.js, Python, FastAPI, Pydantic, DuckDB, Polars, Express, PostgreSQL, Prisma, Redis, WebSockets, Swift, Kotlin, Java, C#, C++, Capacitor, Unity, Unreal, Godot, Cocos, SDKs, CLI tools, MCP, Vercel AI SDK.

Full-Stack Products AI Systems Real-Time Systems Reliability Tooling Eval Integrity Workflow Automation SDK Engineering Mobile Apps Developer Tools

Mobile demo