I am the founder and technical lead of BIDLINE, a voice-to-bid platform for trade contractors, and I am currently available for senior or principal employment with a company that has real operational complexity to build for or automate. Most recently I was Founding Engineer at Octane, where I built an AI-powered back office integration platform for fuel retailers from zero.
BIDLINE is the most complete example of what I build. A plumber, electrician or HVAC tech describes a job out loud on their phone, and moments later the customer has a priced, branded quote they can review, sign and accept. Every price on that quote comes from the contractor's own price book, never from the AI. I designed and built all of it myself: a web application for the business owner, iOS and Android apps for the field, a Super Admin console for running the platform, and the AWS backend, n8n workflows, AI pipeline and operational monitoring underneath. BIDLINE runs under Velocity BPA LLC, the company I use to hold my projects and consulting work.
The node development pipeline I built has published 162 n8n community nodes to npm under the @velocity-bpa org scope. Nine custom MCP servers connect Claude AI directly to Airtable, n8n, GitHub, AWS, Notion, Stripe, npm, Dropbox, and a Claude Agents orchestration layer. I also use n8n as a full CI/CD orchestrator: the 45-node pipeline coordinates parallel Claude AI code generation, GitHub repo creation, automated testing, and npm publication end to end.
Before automation I spent 30-plus years in software development, starting in 1996 at the dawn of the web at companies including AbbVie, CVS/Aetna, Pandora, Samsung, and Apple. I have been building in blockchain and crypto since 2013. Certified in n8n at Level 1 and Level 2. I evaluated Zapier and Make before committing to n8n, specifically because enterprise organizations require the ability to self-host and control their own data environments, and I understood that from three and a half years inside AbbVie.
BIDLINE: Voice-to-Bid Platform for Trade Contractors
2026 - Present | Founder and Technical Lead | bidlineapp.comContractors lose jobs because quoting happens at night, after the work. BIDLINE turns a spoken job description into a priced quote the customer can sign the same day. One person designed and built the whole platform: three product surfaces on one API, with the infrastructure, AI pipeline and operations behind them.
Web Application app.bidlineapp.com | React, Vite, TypeScript
- Back office for the business owner: bids, jobs with a fleet map, calendar, customers, invoices, follow-ups, seasonal campaigns and an intake leads inbox
- A price book with import, markup rules and distributors, plus supplier-invoice capture with line-by-line review that commits received stock to an inventory and cost ledger
- Owner analytics: win rate and pipeline, catalogue and realized margin, cash and receivables aging, crew performance and estimate accuracy
- Workspaces with Owner, Admin, Manager and Worker roles, capability-gated screens, and subscription billing through Stripe
Mobile Apps iOS and Android | React Native, Expo
- Built for the truck: a center voice button records the job, AssemblyAI transcribes it, and the priced bid comes back to the phone ready to review and send
- Jobs, calendar and customers in the field, and on-site customer signatures captured full screen, with a signed PDF copy of every acceptance
- One person can run their own business and work on someone else's crew from the same login, so a worker can grow into an owner on the same app
- Preparing for release on Google Play and the Apple App Store
Super Administration Platform console | role-gated
- A 360 view of every contractor with bids and customers, users and invitations, billing queues, Stripe Connect accounts and card payments, and a read-only view-as mode for support
- Compliance tooling: an audit log with reasons and diffs, a consent log with CSV export, and an account erasure queue
- Runtime control without a deploy: platform settings, feature flags, and an editor for the AI prompt templates with controlled activation
- A Diagnostics console covering services, database, workflows, alarms, usage and cost, including measured AI cost per bid (about two cents)
- AI pipeline: Claude generates the bid from the transcript, Voyage AI embeddings and Qdrant retrieve matching items from the contractor's own price book, and the platform applies authoritative prices so the model can never invent one
- Eight production n8n workflows run customer intake, voice-to-bid, PDF generation, bid sending, notifications, bid indexing, PDF reconciliation and supplier-invoice extraction, with every call to the API HMAC-signed and the payload contracts checked in CI
- Built to be operated by one person: separate staging and production lanes, a synthetic prober that drives every workflow on a schedule, 78 CloudWatch alarms held to a register in CI, and an n8n drift gate that fails the build when live workflows and the repository disagree
Octane: Fuel Retail Integration Platform
Jun 2026 - Jul 2026 | Founding Engineer, Octane (Houston, TX)- Built the integration platform from zero: source-neutral ingestion spine (SourceArtifact, ImportRun, StagingRecord, canonical Postgres) with pluggable file and API transports, a BlobBackend abstraction over local, GCS, and SFTP storage, per-tenant secret rotation against GCP Secret Manager, and a declarative connector manifest system with CI drift and acceptance gates
- Shipped three production connectors end to end: QuickBooks Online (OAuth, CDC incremental sync, Intuit production approval), Clover POS (OAuth2 paged ingestion through canonical promotion), and the Gilbarco/Invenco Passport forecourt XML gateway (Conexxus NAXML v3.2-3.4 classified against a live store's edge capture)
- Drove the Clerk authentication production cutover replacing Google IAP on Cloud Run: shadow-mode canary that caught a row-level-security defect, enforced traffic flip, multi-tenant store onboarding, and invited-guest access with live store-scope delegation
- Hardened the LLM invoice pipeline: human-keyed golden set (6 invoices, 392 lines reconciled to the cent), root-caused a silent-skip validation defect, shipped an agentic edge-case resolver proven in an isolated Daytona sandbox, and delivered a LangSmith-scored accuracy analysis across 20 invoices identifying candidate ranking, not model quality, as the binding constraint
- Defined the platform PII and cardholder-data sanitization doctrine: default-deny per-field treatment, in-memory-only sanitizer, sanitized-only object storage, plus stop-the-bleed remediation after discovering raw cardholder data in the capture bucket
Automated Sales and Client Delivery Pipeline
February 2024 - Present | Velocity BPA- Designed and built an 8-workflow end-to-end automated business system: a prospect submits a form on the website, Claude AI scores and qualifies the lead, extracts API specifications, runs a feasibility analysis, generates a quote, and delivers it via AWS SES, all without human involvement
- On payment via Stripe, the system triggers a build approval gate, routes the job through the 45-node n8n development pipeline, publishes the completed package to npm under @velocity-bpa, and automatically onboards the client by creating an Airtable record, generating a Notion project page, and sending an SES welcome email
- The lead scoring and qualification layer uses Claude AI to analyze inbound requests, assess complexity, and route to the appropriate workflow branch, replacing what would otherwise require a sales and project management function
Operational Intelligence and CEO Agent Suite
March 2026 | Velocity BPA- Built 5 n8n workflows forming the operational intelligence layer: twice-daily Telegram briefings aggregating pending approvals, new client activity, and inbound inquiry classifications generated by Claude Haiku
- Automated invoice chase system: daily Airtable query, Claude Haiku-drafted emails staged as Outlook drafts for one-click sending, full activity log written back to Airtable
- Daily approval processor delivers pending-approval items with a one-tap bulk-approve button via Telegram, reducing the approval function to under a minute per day
- Client Onboarding Enricher fires on Stripe checkout: simultaneously creates the Airtable record, generates a Notion project page, and sends the SES welcome sequence with no manual steps
MCP Server Architecture: 9 Production Servers
February 2024 - Present | Velocity BPA- Designed and deployed nine production MCP servers in TypeScript giving Claude AI direct programmatic control over all business systems: Airtable, n8n, GitHub, AWS (S3, CloudFront, Route 53, SES, Lightsail), Claude Agents (23 tools), Notion (18 tools), Stripe, npm-stats, and Dropbox
- Built an approval-gate architecture where Claude AI executes autonomously on well-defined tasks and automatically escalates ambiguous decisions to a human-in-the-loop Telegram gate
- Deployed OpenClaw, a self-hosted AI agent gateway, on a private server to provide Telegram access to the full MCP tool suite from anywhere, enabling remote autonomous operations without cloud dependency
Node Development Pipeline
February 2024 - Present | Velocity BPA- Built a 45-node n8n workflow that accepts a service name as input and outputs a complete, publishable n8n community node package in 4 to 11 minutes, replacing a process that previously required days of manual work. n8n orchestrates the full CI/CD lifecycle: parallel Claude AI code generation, GitHub repo creation, automated testing via GitHub Actions, and npm publication under @velocity-bpa
- Shipped 162 production n8n community nodes under the @velocity-bpa org scope covering blockchain and DeFi (Uniswap, Aave, XRPL, Solana, Kraken, 80-plus others including AI-native nodes: SingularityNET, Autonolas, ElizaOS, Ocean Protocol, Bittensor, EigenLayer), enterprise SaaS (Auth0, Braze, Amplitude, Segment, Bitwarden), fintech (Dwolla, Square, Klarna, Railsr, Adyen), and healthcare (Epic FHIR, Change Healthcare, Availity)
- Wrote the prompt engineering layer for Claude AI: Sonnet for code generation, Opus for architecture decisions. Added parallel execution with Promise.all so 10 to 20-plus resources are generated simultaneously per run. All Claude API calls use Anthropic prompt caching.
- Implemented a Qdrant vector memory layer (self-hosted on AWS Lightsail) that makes the pipeline self-improving with every build. Before each run, a Voyage AI embedding is generated and queried against a build_memory collection for the top-5 most similar past builds (similarity >= 0.65). Auth patterns, documented gotchas, and resource structures from those past builds are injected directly into Claude's prompts. After each build, metadata is re-embedded and stored back, so every build improves the next.
Claude Conductor: Open Source Desktop App
2024 - Present | Velocity BPA | github.com/Velocity-BPA/claude-conductor- Built a macOS desktop application using Tauri (Rust and Node.js) that solves a real limitation in Claude Desktop: only one MCP server configuration can run at a time. Claude Conductor manages named profiles, each with its own MCP server set, and launches multiple fully isolated Claude Desktop instances in parallel with no shared state
- Credentials and API keys are stored per-profile in the macOS Keychain, never written to disk in plaintext. Running instances display live status with Focus and Kill controls. Profiles can be imported directly from an existing claude_desktop_config.json
- Built to solve a daily operational problem: switching between client contexts, dev and production configurations, and different MCP tool sets without manual config editing and app restarts
API Change Monitor
2024 - Present | Velocity BPA- Three-tier monitoring system watching 162 published integrations: OpenAPI spec diffing, MD5 documentation hashing, and GitHub release tracking
- Claude AI scores detected changes 0 to 100. Anything scoring 60 or higher triggers an automatic rebuild, keeping the entire catalog current without manual initiation
Cross-DEX Arbitrage Bot Suite
March 2026 | R&D / Portfolio Project- Eight-workflow n8n system monitoring price spreads across Uniswap, SushiSwap, 1inch, OKX DEX, and Kraken on a 30-second polling interval, with paper trading tracked across nine Airtable tables
- Gas optimization engine querying Etherscan Gas API every 15 minutes to compute dynamic minimum-spread thresholds
- ML threshold optimizer processing historical data across three strategy profiles mapped to market volatility. Converges near 1.5% spread at full training confidence
- Portfolio risk manager running every five minutes, plus a Bootstrap and Chart.js real-time dashboard on 30-second refresh cycles
Enterprise Treasury Automation
March 2026 | R&D / Portfolio Project- Eight n8n workflows integrating SAP, GTreasury, Workday, BlackLine, and Coupa with XRPL blockchain settlement rails: cross-border AP payments, liquidity sweeps, milestone escrow, FX hedging, intercompany netting, automated reconciliation. Based on exploratory R&D conducted with AbbVie treasury and finance stakeholders and calls with Ripple exploring ODL cross-border payment integration.
- Each workflow applies configurable approval-gate logic before executing payment instructions and persists full audit trails to Airtable for compliance review
Client: Hello Crypto
2024 | Sao Paulo, Brazil- DEX Opportunity Tracker dashboards and AI-powered trading bots for a blockchain startup using n8n workflows and Bootstrap
- Real-time market analysis algorithms using OpenAI GPT and Anthropic Claude APIs with DEX integration
- n8n Course Level 1 and Level 22025
- Decentralized Finance SpecializationDuke University, 2023-2024
- Neural Networks and Deep LearningDeepLearning.ai, 2024
- Generative AI with Large Language ModelsDeepLearning.ai, 2023
- Adobe Experience Manager: Sites Front-End and Back-End DeveloperAdobe, 2022
- Business of MusicNew York University, 2001