Selected Work

AI agents, automation engines, and production delivery.

Each project is framed by business problem, implementation detail, and operational result.

01 / Featured system · Platform

OpenClaw / Dispatch Router

The challenge
Personal AI tooling was fragmented, requiring manual context-switching across 25+ disparate agents and workflows.
What I built
Architected a personal multi-agent orchestration platform with a dedicated Dispatch routing layer, enforcing quality gates, isolated sessions, and zero-token-waste background automation.
Result
One self-hosted system handling continuous research, GitHub webhook alerts, automated code review, and content production — fully owned, no NDA required.
Anthropic SDKOpenRouterKilo CodeGitHub ActionsLocal Ollama
Dispatch routing field

Signals in

  • Continuous research
  • GitHub webhooks
  • Agent requests

Control plane

Dispatch Router

Route · isolate · gate

Fully owned

Governed work out

  • Automated code review
  • Content production
  • Webhook alerts

25+

Active agents

0

Token-waste baseline

PhaseDesign → Deliver
25+ active agents routed through isolated sessions, shared quality gates, and zero-token-waste background automation.

Project ledger

Problem. Build. Result.

Filter the record by project type, then open any case study for the implementation details and operating evidence.

Filter

Showing 5 of 5

AI Agent

Transition Readiness Agent

8+

systems unified

Glean AgentSOQLSalesforceGongTotango+1 more
Challenge
CSMs inherited accounts with no compiled context after onboarding handoff.
What I built
Built a Glean Agent synthesizing 8+ systems — Salesforce, Gong, Totango, Provisio, Zendesk, Gmail, Drive, Slack — into a ready-made transition package with CSM talking points.
Result
CSMs get full customer intelligence compiled automatically, no manual reconstruction needed.

AI Agent

REACH Scoring Agent

$250M

ARR under scoring

CSMs get scored accounts delivered to Slack, review in Glean, and push confirmed scores to Totango in one click.

Challenge
Account prioritization depended on gut feel — no standardized scoring across 1,200+ accounts.
What I built
Built a Glean Agent scoring accounts on a 90-day cadence and key events, with Slack notifications, human review, and one-click Totango import with historical tracking.
Glean AgentSalesforceTotangoProvisioGong+2 more

Algorithm

CSM Assignment Engine

1,200+

accounts balanced

Google Apps ScriptTotango APISalesforce APIGlean APIConstrained Greedy Heuristic+1 more
Challenge
Assigning accounts purely on headcount count creates hidden imbalance — a $50K single-product account and a $500K five-product account are treated identically, overloading CSMs with lower headcount but higher complexity.
What I built
Designed a Constrained Greedy Heuristic that evaluates four dimensions (Volume 35%, ARR 25%, Renewal Balance 25%, Portfolio Risk 15%) with dynamic capacity re-calculation. Includes the Bouncer pattern for white-glove Enterprise handling, ARR anchoring (vs TCV) to avoid renewal penalties, and a New Hire Magnet for accelerated ramp.
Result
Dynamic, complexity-weighted assignment replacing manual spreadsheet passes — 1,200+ accounts balanced across the team with zero manual reconciliation cycles.

Internal App

Portfolio Redistribution App

Days→Min

workflow compression

Territory planning that took days now takes minutes, with scenario comparison and handoff tracking.

Challenge
Redistributing portfolios for territory changes was a multi-day manual process with no scenario modeling.
What I built
Shipped an internal planning app for ARR redistribution scenarios and handoff planning — reducing multi-day workflows to minutes.
Google Apps ScriptGWS APITotangoScenario ModelingGreedy Assignment

Orchestration

Multi-Agent Design Sprint

0

merge conflicts

Claude CodeGit WorktreesNext.js 16Playwrightaxe-core
Challenge
A full-site redesign hit list spanning seven pages — and parallel coding agents collide on a shared codebase.
What I built
Fanned five coding agents into isolated git worktrees with file-disjoint scopes, governed by one committed design brief — three named primitives plus hard bans — with lint/typecheck gates per agent and a single integration pass.
Result
Seven pages restyled in one session, five branches merged with zero conflicts — and this site is the artifact.