Backend Engineer • AI Infrastructure
Production AI infrastructure. I built and ship Festival, the camp and fest CLIs that organize agent workflows. The stack is in daily use coordinating 27+ projects in one shared execution environment, with 90%+ token reduction versus unstructured agent runs. I also wrote claude-code-go, the first full-featured Go SDK for Anthropic's Claude Code, shipped within days of API availability.
Open-source Go developer tooling. Beyond Festival and Claude Code Go I've released a steady stream of small Go tools for working with AI and code: tcount (token counter for files and directories, also wired into a Neovim plugin), tree2scaffold (build project layouts from an ASCII tree), gomod-rename (refactor Go module paths cleanly), and algo-scales (open-source CLI for algorithm interview prep with AI tutoring, plus a companion Vim plugin).
Multi-agent and agentic systems. At ETHDenver 2026 I shipped the Obey Agent Economy: five autonomous agents coordinating across Hedera, 0G, and Base. Includes a Go inference agent on 0G's decentralized GPU network with ERC-7857 iNFTs, an autonomous DeFi agent on Base, settlement contracts, and a co-authored HIP-1215 ReputationDecay standard for agent reputation.
Backend at scale. I build production Go and Python backends that handle real load: OTC crypto trading platforms with KYC and compliance workflows, NFT marketplaces handling 15M+ on-chain transactions per month (at Mythical Games) with 100% finality, and custodial wallet services that let non-crypto-native users hold and trade assets without ever seeing a seed phrase.
Modernization in high-control environments. Inside Bank of America's Global Security group I rebuilt a 15-year-old codebase and shipped a CI/CD pipeline that cut deploy cycles from months to hours, reclaiming roughly six developer-months of productivity every year inside one of the most restricted enterprise environments in finance.
1. Plan first, code once. I spend real time on planning so the v1 deliverable is the right thing built well, not a fast first draft your team has to throw away. Most of the AI tooling I write enforces the same discipline on agents.
2. Systems thinking. I trace features to revenue, latency to user experience, and compliance to risk. That map guides architecture decisions long before any code gets written.
3. Knowledge transfer is part of the job. Clean commits, working docs, and team walkthroughs ship with every engagement. The goal is for your team to keep moving after I leave, not to keep me on retainer.
If your team is paying for AI tools but output is inconsistent, or you have hard backend, blockchain, or AI-infrastructure problems and want a senior engineer who has shipped them before, send a note. I bring engineers, not slide decks.
Public release of the camp and fest CLIs: modular tools for AI workspace management and hierarchical autonomous execution. Users own their data and can swap underlying AI providers without lock-in. Designed around the principle that agents should plan before they execute, track progress as they go, and validate their own work afterward, with correction loops that force them to verify.
fest enforces planning phases before agents start work; goals defined explicitly before any task runs
Impact: Agents know what they're supposed to ship before they start typing, so there are fewer wasted runs and fewer rework loops
Phase / sequence / task hierarchy gives long-running agent work a structure that survives context resets and session boundaries
Impact: Multi-day or multi-session work stays coherent across agent invocations
Quality gates at the end of every implementation sequence (testing, review, iteration) that force agents to verify their own work before declaring it done
Impact: Catches the failure mode where agents mark tasks complete without actually shipping the work
Modular CLIs (camp + fest) with Claude Code plugin distribution. Users own their workspace data and can swap AI providers without lock-in
Impact: Works against any AI provider; user data never leaves their environment
ETHDenver 2026 stack: a working multi-chain AI agent economy where five specialized agents coordinate across Hedera, 0G, and Base. Demonstrates production-grade agentic architecture: on-chain identity, decentralized inference, autonomous DeFi, observability, and orchestration, all built as open source.
Go orchestrator dispatching tasks via Hedera HCS, managing HTS payments and quality gates across the agent fleet
Impact: Single source of truth for cross-chain task dispatch and settlement
Go AI inference agent running on 0G's decentralized GPU network, with ERC-7857 iNFTs for provenance and 0G Storage/DA for the audit trail
Impact: Decentralized inference with cryptographic provenance, no centralized model black box
Autonomous Go DeFi agent on Base. Uniswap V3 trading, ERC-8004 identity, x402 payments, ERC-8021 attribution
Impact: Live autonomous trading with full on-chain identity and attribution
Solidity contracts: AgentSettlement, ReputationDecay (HIP-1215 author proposal), ERC-7857 AgentINFT. Full Foundry test suite (34 tests)
Impact: On-chain primitives for agent reputation and settlement; co-author on the HIP-1215 reputation-decay standard
Next.js real-time observer dashboard with 5 live panels: Festival View, HCS Feed, Agent Activity, DeFi P&L, Inference Metrics
Impact: End-to-end visibility across the multi-chain agent economy in one interface
Hiero CLI plugin for camp workspace management with five scaffold templates for Hedera and 0G development
Impact: Reusable scaffolding for the next generation of Hedera + 0G agent builders
Risk-management primitive for autonomous DeFi agents. Eight sequential risk gates evaluate every agent action; each decision is recorded as an immutable receipt via Chainlink's Compute Runtime Environment. Solves the 'how do I trust an autonomous agent with capital' problem with verifiable, auditable risk gating.
Layered risk evaluation pipeline that every agent transaction must pass before execution
Impact: Failure modes are caught at gate boundaries with clear attribution rather than at runtime with surprise losses
Each risk decision recorded immutably via Chainlink Compute Runtime Environment for after-the-fact auditing
Impact: Verifiable audit trail: autonomous agent actions become accountable, not opaque
Production multi-agent chat platform built on top of the Guild orchestration architecture. YAML-based prompt routing with per-agent context windows; demonstrates Guild's primitives in a shipped product.
Declarative prompt routing config: prompts and agents wired together via YAML with per-agent context windows
Impact: Non-engineers can configure agent routing without touching code
Each agent maintains its own context state and history; conversations route between agents based on intent and capability
Impact: Agents stay within their lane; less context pollution; cleaner conversational handoffs
Shareable context system for teams: persistent memory and knowledge sharing across LLM sessions. Solves the 'every agent starts from scratch' problem by giving teams a shared knowledge layer that any agent can read from.
Shared knowledge layer that persists across sessions and agents; each new conversation starts already knowing the team's facts and decisions
Impact: Teams stop re-explaining the same context to every new chat session
Public source-available collaborative AI agent framework, written in Go, designed around the question: how does a single human direct hundreds or thousands of specialized agents toward one multi-faceted goal? The core orchestration primitives (agent definitions, gRPC inter-agent communication, SQLite + vector memory, multi-layer prompt management, hierarchical 'managers of managers' coordination, and kanban-style task review) are working and underpin the camp/fest CLI ecosystem. Public R&D status: pre-MVP overall; the parts the camp/fest CLIs depend on are battle-tested in daily use.
Specialized agents coordinate via gRPC with task dependencies and a managers-of-managers hierarchy. A single human can direct large agent teams without micromanaging each one
Impact: A working primitive for the 'CEO of a 24/7 agent team' problem
Persistent memory across sessions with vector search; agents share context across runs without re-explaining everything every time
Impact: Agent context survives session boundaries, process restarts, and handoffs between agents
Composable prompt system with dynamic context injection and template management; the same prompt patterns work consistently across model providers
Impact: Behavior tuning without rebuilding agent code; provider-agnostic by construction
Every agent action is planned before execution and tracked against the plan; quality gates at the end of each implementation sequence force agents to verify their own work and surface mistakes early
Impact: Catches the 'agent marked it done but didn't actually ship it' failure mode before mistakes compound
Go SDK providing a native interface to Claude Code CLI features, enabling developers to build AI-powered applications with Claude's coding capabilities.
Complete Claude Code CLI wrapper with streaming support and context cancellation
Impact: Native Go interface for Claude Code features
Model Context Protocol (MCP) integration with multiple output formats (text, JSON, streaming JSON)
Impact: Advanced AI tool integration capabilities
Multi-turn conversation support with file processing capabilities
Impact: Production-ready conversational AI features
Leverages Claude Max subscription to avoid API costs
Impact: Reduced AI development costs for Go developers
CLI tool for summarizing YouTube video transcripts using Ollama for 100% local processing. Supports intelligent transcript extraction, interactive TUI, and research planning features.
Uses Ollama for 100% local processing without external API keys
Impact: Complete privacy with no external data sharing
Prefers manual captions, supports multiple input formats (.txt, .list, .urls, .csv)
Impact: Flexible input handling with intelligent caption selection
Create focused research plans to extract targeted content across multiple videos
Impact: Comprehensive research document generation from video corpus
Video chunk processing, transcript caching, comprehensive logging
Impact: Efficient processing of long videos with persistent cache
Early-stage open-source algorithm learning tool for coding interview preparation. Built in Go with a terminal-first design to help developers practice 11 fundamental algorithm patterns.
Organizes coding problems by algorithm patterns (Sliding Window, Two Pointers, etc.) similar to musicians practicing scales
Impact: Structured approach to algorithm pattern recognition
Learn, Practice, and Cram modes with progress tracking and daily practice workflow
Impact: Flexible learning experience with built-in accountability
AI-powered hints and explanations using Ollama or Claude Code via custom Go SDK
Impact: Enhanced learning with AI assistance
Vim plugin for in-editor problem solving, with Neovim and VS Code support planned
Impact: Practice within familiar development environment
Automated my job by building a neural network using python and numpy to write regular expressions and SQL code based on business requirements.
Generate regular expressions and SQL code based on business requirements
Impact: Reduced workforce requirements
Automated an incredibly tedious business process by building a tool that performed a 3 month excel based analysis in 30 seconds with a button click
Impact: Reduced workforce requirements
Applied machine learning to robotic warehouse debugging and maintenance scheduling.
ML models for warehouse efficiency optimization
Impact: Improved maintenance planning, reduced firedrills and system shutdowns
Integrated ML predictions into warehouse automation systems
Impact: Smarter warehouse operations
Web3 first-person extraction shooter built on GalaChain. Engaged via Blockhead Consulting to own the blockchain backend and custodial wallet service that bridges the Shrapnel game client and the chain that underpins its player-owned economy.
Built the custodial wallet service that lets non-crypto-native players hold and trade in-game assets without managing private keys themselves
Impact: Removed the wallet-onboarding friction that kills web3 game adoption. Players never see a seed phrase
Backend services bridging the Shrapnel game client and GalaChain: transaction signing, asset state, on-chain settlement for a AAA FPS title
Impact: Production blockchain infrastructure for a shipping AAA web3 game
Created local Hyperledger Fabric tooling so blockchain microservices could be exercised against a real-feeling chain in dev and CI without standing up a shared testnet for every engineer
Impact: Accelerated microservice testing turnaround and removed shared-testnet contention from the dev loop
Led backend development of a white-glove OTC crypto trading platform serving community banks and credit unions. The platform processed $20M+ in trades end-to-end, including KYC verification, trade execution, and the compliance workflow regulated buyers require.
Django backend powering the white-glove OTC trade execution path; handled order intake, KYC verification, trade settlement, and the compliance audit trail
Impact: Platform processed $20M+ in trades for regulated financial-institution buyers
Wired the trading backend into Zoho CRM and built internal dashboards so the operations team could see deal flow, KYC state, and exception cases in one place
Impact: Operations team had a single source of truth for every trade and KYC artifact
Built KYC verification flows and automated fraud-prevention checks against the OTC trade pipeline
Impact: Caught suspect trades before settlement; kept the platform inside the compliance envelope its regulated buyers required
Blockchain infrastructure development for gaming-focused NFT marketplace handling millions in transactions per day.
Golang microservices handling on-chain transactions
Impact: 15M+ transactions/month with 100% finality
Solidity contracts for gaming and marketplace features
Impact: Audited contracts deployed to production
NFT bridge architecture to mainnet Ethereum
Impact: Enabled asset portability between networks
Developed dApps and blockchain infrastructure for enterprise blockchain platform.
Python/Node.js microservices for rapid dApp prototypes and POCs
Impact: Accelerated client project delivery
Architected platform with weighted voting using time-held coins
Impact: Novel governance mechanism implementation
Smart contracts and Go backend for decentralized social media
Impact: Production social media dApp
NFT data intelligence tool built at EthDenver. Aggregated NFT marketplace data and surfaced patterns about collections, holders, and sale velocity in a single interface. Recognized at the hackathon as a 3rd place community winner.
Pulled and normalized data across NFT marketplaces and on-chain sources to give holders and traders a single intelligence view
Impact: Surfaced collection-level signals that were otherwise scattered across marketplaces
Personal project building distributed exchange for physical precious metals trading with DeFi integration.
Distributed system for peer-to-peer precious metals trading
Impact: Enabled physical asset liquidity access
On-site verification system for physical asset trades
Impact: Trustless trading of physical assets
Bridge to use physical metals in DeFi protocols
Impact: Novel asset class for DeFi
Designed and developed a subnet blockchain network for automating complex asset transfers with signficant regulatory documentation requirements.
Consortium subnets chose the best contract engine for their particular use case.
Impact: No reliance on niche developers for implementation.
Founded and led major cryptocurrency meetup group in Charlotte, North Carolina.
Led modernization efforts for one of America's largest financial institutions, focusing on enterprise CI/CD infrastructure and legacy security scanning systems.
Designed and implemented CI/CD pipeline for 15-year-old codebase
Impact: Reduced deploy times from months to hours
Modernized legacy Python security scanning software
Impact: Ensured compliance with modern secure coding standards
Automation and process improvements
Impact: Reclaimed 6+ months/year of lost developer time
Conducted comprehensive audit of data processing pipelines for major regional bank.
Comprehensive security and accuracy audit of Python data pipelines
Impact: Found coding errors missed in an OCC remediation effort affecting thousands of customers
Performance optimization of data processing systems for payment processing company.
Performance tuning and cost optimization of ETL processes
Impact: Cut AWS spend by $40K/month
Worked on order processing systems for one of the largest photo commerce platforms.
Enhanced Python-based order workflow engine routing online orders to factory processes
Impact: Improved order processing reliability and throughput
Assisted in migration planning of Forge from on-prem servers to AWS
Impact: Modernized infrastructure for better scalability
Nine months of running my entire workload through Festival: 138 repositories in git-tracked campaigns, 26.5 billion tokens at 95.8% cache reads, and overhead that stayed flat as the work scaled.
A battle-tested workflow for using Claude Code to manage large-scale projects with parallel AI agents working like a distributed engineering team
I delivered $1 million worth of software per week for seven straight weeks using AI. Traditional COCOMO values the 219,400 lines of executable code at \(7.8M. My actual cost: \)636 in AI subscriptions. This is how I did it.
Let's discuss how my experience can help accelerate your project.