AI & Automation lead at Atlas HXM. I design, build, and harden the systems that let 50+ people run a global GTM org on AI — and I ship the frameworks that let others do the same.
ROSTR in action — multi-agent architecture for production AI. Explore the full framework →
A unified architecture for production-grade multi-agent systems — published, cited, and running in production. Four components, a five-phase taxonomy, and a persistent knowledge architecture that powers agent systems at scale.
📄 DOI: 10.5281/zenodo.19550414 ↗The Intent Compiler that transforms loose natural language into strict agent runtime manifests. Five-stage pipeline: intent extraction → context injection → semantic enhancement → runtime compilation → output routing. Takes "improve the outreach workflow" and produces a fully-parameterized agent manifest with model, temperature, tool permissions, memory mode, and completion criteria. No more raw prompts — every agent invocation flows through compiled manifests.
Autonomous multi-pass retrieval with source credibility stratification. Three-tier architecture: Tier 1 (academic/authoritative, credibility 1.0), Tier 2 (editorial/verified, 0.75), Tier 3 (community/UGC, 0.40). Multi-pass algorithm: broad sweep → gap fill → deep verification. Confidence scoring formula weights source count, consistency, tier distribution, and recency. Knowledge is ingested, embedded, and compounded across sessions.
A 5D phase taxonomy classifies every task: PreD (research), Design (architecture), Development (build), Deployment (ship), Debugging (fix). Four-dimensional priority scoring weights phase urgency, dependency impact, business impact, and resource efficiency. Agent allocation uses vector similarity + specialization overlap + load scoring. Orchestration patterns: sequential chains, parallel fan-outs, aggregation fan-ins, conditional branches.
Persistent reference architecture with four-level state management: session (ephemeral), project (persistent — decisions, artifacts, learnings), organization (evolving — identity, ICP, positioning), agent (portable — skills, preferences, calibration). Knowledge compounds across sessions. The Hub provides scoped namespaces, hierarchical credibility, and cross-namespace permission gates. Agents don't restart from scratch — they load context from the Hub.
The ROSTR framework powers the CoE project intake system — cutting idea-to-scoped-project from 4–8 hours to under 30 minutes. Every GTM automation project routes through PAL compilation and NPAO phase classification.
The agency's core product: SOP → intelligent agent. JTBD Analysis maps intent, PAL/NLP extracts and compiles it, the ROSTR system builds the agent. Every client engagement runs on this pipeline.
Full academic paper on Zenodo with DOI. Formalized architecture with algorithms, formulas, and implementation templates. Not a blog post — peer-reviewable, citable research. Read the paper ↗
Production-grade tools across AI, orchestration, GTM, and cloud.
Production AI systems shipping inside a 50+ person GTM org at Atlas HXM.
46-node n8n system auto-screens pipeline deals against 30-sector compliance worksheet. Hybrid LLM + deterministic engine.
Recovered failing multi-agent follow-up pipeline. Replaced dead AI Search with deterministic blob lookups. 882KB → 295KB context.
Meeting booked → prep note to rep within hours. Event trigger defeats time-zone failures. Factors.ai enrichment.
AI classify-route + CRM attribution. Recovered 364 historical contacts. Idempotency-gated pipeline.
Traced 515K credit spike to root cause. Built enrichment template with ICP gate, deduped phone model, 3 build profiles.
Full supply chain for outsourced SDRs: intake → enrich → suppress → activate. 5,398-contact master, CRM-safe.
30+ reusable Claude skills and agents for the GTM team — prospecting, workflow ops, revenue enablement, content at scale.
30+ skills total. Full catalog in master knowledge base.
Products and platforms I've founded, built, and shipped — from AI consulting to civic infrastructure.
Production AI for SMB and mid-market: chatbots, email automation, ed-tech, Shopify workflows.
gptpat.co →Civic intelligence platform for Chicago's 48th Ward — precinct directory, business database, intake forms, ward dashboard.
your48.vercel.app →GTM systems consulting. Dark Roast Media ($1M→$3M ARR), Rand McNally ($5M pipeline, 2,000% engagement lift).
revenabled.com →Published multi-agent architecture: PAL + RAG DAL + NPAO + Rostr Hub. 5D phase taxonomy. Zenodo DOI.
Zenodo DOI →Electronic Press Kit platform for independent artists. Dark theme, gold accent. Next.js 16 + Supabase + Stripe.
artistsepks.vercel.app →GTM systems for private-AI / federated-data startup. ICP, outbound tooling, CRM automations with PhDs.
scalytics.io →Agent development & adoption agency. SOP → intelligent agent via ROSTR. AI teams, n8n/Make, IDE setup, custom MCPs.
gencyai.vercel.app →I'm a production AI systems builder based in Chicago. I lead AI & Automation for a global Employer-of-Record platform, where I've shipped 8+ production automations that screen deals, prepare reps, automate follow-ups, and govern spend — all built on a hybrid architecture where LLMs handle the messy text and deterministic code makes every decision auditable.
Before AI, I spent years in enterprise sales at Uber and Upwork, which is why I build differently: every system I ship is designed for the operators who will actually use it. I also serve as a precinct captain for Chicago's 48th Ward, where I built a civic intelligence platform connecting residents to their government.
Complete FDE / GTM / AI Engineer resume with experience, civic leadership, publications, and impact metrics.
View Full Resume