Meeting reminders
Writes a personalized reminder email and text for every booked meeting.
GTM engineerRevenue systemsAI agents
Salesforce, the website, the ad platforms, and the AI agents that run between them. I turn sales, marketing and onboarding into one measurable system, then put AI to work inside it.
Taking one outside build at a time
Most go-to-market stacks are six tools that don't talk to each other. I build the connective tissue, in the CRM, on the website and inside the ad platforms, and I put agents on the work nobody has time for.
Ads and search
Site, pages and forms
Booked to held
Pipeline and follow-up
Kickoff to live
Risk, renewal, win-back
Each one has the problem, what I built, the result, and the method behind the number. All of it shipped in-house at a venture-backed B2B SaaS company.
Personal first-touch emails to every new lead, and win-back sequences to lost deals and churned accounts.
New leads got a generic sequence, and lost deals and churned accounts got nothing at all.
AI email lanes that sit on top of the sales engagement tool. One writes the first personal email to every new lead who hasn't booked. Others write win-back outreach from the account's own history: why they left, what they used, what has shipped since. Every send is logged with replies and outcomes, and variants are assigned at random so results can be trusted.
ReceiptReply rate is an early read against the previous sequence. Deals counted within 75 days of enrollment.
AI reminders, texts that read replies, and automatic slot refills on every booked sales meeting.
Booked meetings were no-showing or cancelling at the last minute, and the empty slots were wasted rep time.
An engine that writes a personalized reminder for each prospect, sends morning-of emails and texts, reads the replies, and puts freed slots back on the calendar. The AI copy was tested against a generic version on a random split.
ReceiptMeasured on a randomized split of 2,266 meetings.
Reads sales calls and writes the answers into the CRM.
Reps left CRM fields blank, so forecasting, routing and reporting ran on missing data.
An agent that reads every sales call and writes what it learned into the CRM, with guardrails so it can only touch fields that have been approved.
ReceiptHours are counted as if done by hand, at five seconds a field.
Scores every booking for risk and frees rep calendars hours earlier.
Reps were spending prime calendar time on meetings that were never going to happen: bad-fit leads, duplicates and no-shows.
A scoring engine with a rule layer plus an AI auditor. It scores each booking, explains its reasoning, and recommends which meetings to cancel so the slot can be reused.
ReceiptOutcomes counted only for meetings left on the calendar.
A CRM home page that tells each rep who to call next, backed by one ledger of every customer touch.
Speed-to-lead and follow-up weren't measured anywhere. Activity lived in separate tools, and reps had no single list of what to do next.
An operating layer for sales inside the CRM: one ledger fed by every dialer, sequencer and text, speed-to-lead clocks, a recovery path for no-shows, and a home page with one prioritized list and one-click outcomes.
ReceiptLedger reconciled record by record against the source systems.
Closed revenue flows back to the ad platforms, and a monitor catches spend chasing the wrong signal.
The ad platforms were bidding on page views and form fills, not on meetings held or deals won, and cost per lead had doubled without anyone knowing why.
An engine in the CRM that sends meetings held and deals won back to every ad platform, with delivery checks so nothing is silently dropped. Alongside it, a full audit of the search account that traced the cost increase to its cause, and a monitor that flags when bidding starts chasing a low-intent signal.
ReceiptA finding and a fix, not recovered dollars. No bidding lift is claimed yet.
A shorter form, faster landing pages, and automated tests on every release.
The lead form was slow and three steps long, landing pages were heavy, and nothing checked the funnel before a change went live.
A one-step form straight into the scheduler, landing pages rebuilt for speed, attribution that survives cookie blockers, and a browser test suite that blocks any release that would break the path from ad click to booked meeting.
ReceiptNo conversion lift is claimed. None has been measured yet.
One AI-readable map of every GTM system, with an agent on top that answers questions and ships changes safely.
Every data question and every small change needed an engineer, and nobody had one view of how the stack fits together.
A living catalog of every field, flow and workflow across the stack, a glossary that maps business terms to exact fields, and an AI teammate on top. Anyone can ask it a question. Builders can have it ship changes, with approvals, backups and limits on what it may touch.
ReceiptCounts come from the synced catalog, not estimates.
When a tool's job can be done better in the CRM or by an agent, I build it and migrate cleanly.
The stack had grown one tool at a time. Data was split across systems, renewals kept arriving, and several tools did jobs the CRM or an agent could do better.
A customer success workspace inside the CRM, sales sequences with AI steps, and an agent that took over list imports, dedupe and enrichment. Each one started with an audit of what the old tool really did, so only what mattered was rebuilt.
ReceiptReplaced or avoided: some tools are already off, some retire at renewal, some were never bought.
Agents that analyze customer calls, update the CRM, send recaps and capture churn signals.
Customer sessions weren't being logged, recaps were written by hand, and churn signals sat in chat threads.
A set of agents for the onboarding and support teams, built with a teammate: one reads each customer call and updates the CRM, one sends the recap, one captures cancellation requests from chat, and one prepares the first call.
ReceiptBuilt with a teammate on a shared agent platform.
Version control, tests, reviews and monitoring for the revenue stack, so changes are fast and nothing breaks quietly.
Changes to the CRM and the marketing stack went out by hand, with no review, no tests and no record of what changed when a metric moved.
A release pipeline for the CRM with test gates, an AI code reviewer and human approval for production. A monitoring service that watches the website, ads and booking flow for changes. And production errors that turn into tickets on their own.
ReceiptAI writes and validates in a sandbox. A human approves every production change.
They read calls, write CRM data, personalize every reminder, score risk and draft follow-ups. Each one is registered, costed per run, and measured against a control wherever a control is possible.
Writes a personalized reminder email and text for every booked meeting.
Texts prospects before meetings and reads the replies, so reps know who can't make it.
Scores each booking for risk and recommends which to cancel.
Reads sales calls and fills approved CRM fields.
Writes the first personal email to every new lead who hasn't booked.
AI-written outreach to lost deals and churned accounts.
Drafts the follow-up from the call transcript and CRM data.
Reads each customer call and updates the CRM.
Sends the customer a recap after each session.
Reads cancellation requests and logs the reason and outcome.
Builds the deck before a customer's first call.
Account-aware customer emails written by AI, tested in shadow first.
Reads appointment emails and books the meeting automatically.
Drafts replies in the sender's own voice, behind an approval step.
Explains every system release in plain English.
Every agent number I publish goes through the same four tests.
Where I can, I run randomized splits against a control group. Numbers that don't survive review get withdrawn, and five were. That discipline is why the ones on this page hold up in front of a skeptical CFO.
Most go-to-market stacks lose money quietly: ad spend that optimizes for the wrong thing, data that stops syncing, leads nobody sees. I find the leak, prove it with data, and fix it.
Most stacks pay for tools whose job the CRM or an AI agent can now do better. I rebuild that job inside the systems you already own, migrate the data cleanly, and you stop paying the renewal.
I work in the tools themselves and in the code around them: Apex and Lightning, TypeScript and Python, CI pipelines and agent tooling.
1,100+ custom fields, 250+ Apex classes, 100+ flows and 39 Lightning components shipped.
Marketing automation, nurture and segment workflows, and a full property catalog for AI.
Claude and GPT in production, MCP servers, tracing, and cost logged per agent run.
The engines behind the AI programs. Production errors become tickets automatically.
Sequencers, dialers and SMS, deduplicated into one ledger of every touch.
Google, Meta, Microsoft, X and ChatGPT Ads, with attribution reconciled to the leadership dashboard.
Webflow, Next.js and Calendly: landing pages, lead forms, routing and scheduling.
PostHog with identity that survives cookie blockers, A/B tests and funnel dashboards.
Intercom, NPS routing, risk scoring, and a two-way bridge into the CRM.
Call transcripts turned into CRM data, and a nightly warehouse sync onto every account.
Apollo and Clay for enrichment, with imports, dedupe and tagging handled by an agent.
GitHub Actions, AWS and Playwright: release gates, monitoring and automated browser tests.
“The texts that are going out on our behalf have been super clutch and have saved me plenty of time, effort, and peace of mind.”
“Wow dude this is INSANE! Really great job, everyone is talking about it now.”
“That's absolutely amazing.”
“Huge lift!”
“Love that you asked for explicit approval. This is approved.”
09 / Work with me
I run GTM engineering in-house and take one outside build at a time. The first build ships in about three weeks at a fixed price, and your team owns it afterwards.
You show me the system that's leaking.
In writing, before anything starts.
Built, tested and in production.
Your team owns it from day one.
Builds I take on
Win-back, first-touch and follow-up emails written from each account's own history, logged and measured.
Reads calls, fills fields, enriches records or scores risk, with guardrails and a measured result.
Reminders, texts and risk scoring on every booking, with empty slots refilled automatically.
Won deals sent back to the ad platforms, so bidding stops chasing form fills.
Speed-to-lead clocks, a single ledger of every touch, and a home page that says who to call next.
The job a vendor does, rebuilt inside the systems you already own, with a clean migration.