Real tools,
doing real work.
Every build below took a job that ate someone's week and put it on autopilot: assembling reports, reviewing sales calls, running the ads. A full day of work becomes an hour. Answers that used to wait days show up in minutes. That's the gap between AI as a junior you re-train every morning and AI as a senior who just needs the goal.
AI Meeting Coach
Every sales meeting scored and coached, automatically.
A five-person sales team with no scalable way to coach every call. The off-the-shelf answer, Gong-class conversational intelligence, runs $20–30K a year, every year, and scores everyone against the same generic criteria.
When a Zoom recording lands in the CRM, the coach picks it up, transcribes it, and scores it against the company's own rubric for that meeting type. Minutes after the call, the rep gets a Slack DM: what went well, what to sharpen, which objections came up and whether they were resolved, and the next step to take. Managers see per-rep trends over time. Marketing sees every objection raised across every call.
The rubric is the company's, not a vendor's: discovery, demo, and proposal calls each score on their own criteria, and the criteria can change as the team learns.
“It's a conversational intelligence tool, very similar to what you'd get with Gong. But we built it all in-house, on tools the team already had.”
Built withClaude Code, the company's CRM and Zoom, Deepgram for transcription, Claude for the analysis, Slack for delivery. Shipped in a month.
AI Business Review Deck Generator
A key account's season review, 18 branded slides, from one typed line.
Every key account's mid-season review took a customer success manager ten hours, sometimes more: pull the season's data by hand, tabulate it, build the slides one by one.
The rep types one line: build a review for this account. The tool pulls the account's season-to-date data from the company's own API, runs the metrics, and flags the items worth discussing with the customer. Then it pauses while the rep captures a few screenshots (the judgment call stays human), assembles the branded 18-slide deck, and lands it in Google Drive.
“What would traditionally take ten hours or more was done in a matter of minutes.”
Built withClaude Code skills orchestrating custom code for the data pull and metric math, the client's own API as the source (real numbers, never retyped), and Google Drive for delivery.
Monthly Reporting Engine
The monthly exec report, assembled by conversation.
The monthly report the exec team was owed took the marketing lead a full day: channel numbers reconciled against the CRM by hand (they regularly disagreed), attribution logic re-derived from memory each cycle, and the deck rebuilt slide by slide.
Two skills for Claude Desktop. No app, no server, no new login. The first pulls marketing-sourced deals straight out of HubSpot, reads the leads export, runs the channel attribution math, and runs six data-quality checks before any number reaches a slide. The second builds the finished 8-slide exec deck in the company's design and editorial voice. The methodology is encoded once, so every report uses the same math instead of being rebuilt from memory.
The marketing exec stays in the loop where it matters: reviewing the hygiene flags, adding the context the CRM doesn't have, final touch-ups.
“A full day of pulling numbers and building slides, down to about an hour. And every report runs on the same math.”
Built withtwo Claude skills installed in Claude Desktop, the HubSpot connector, and the company's reporting methodology written down once. The tool is the AI assistant the team already has, taught how the business measures itself.
Meta Ads Manager
Paid ads answered for in pipeline dollars, not clicks.
A contractor managed the ads. He was good, and he cost 10% of monthly ad spend, every month. Worse: nothing connected Meta's numbers to what happened after the lead form. A campaign full of cheap leads that never became pipeline still looked like a winner.
A small toolkit driven through Claude Code in plain English. Ask for the weekly report and it pulls last week's performance ad by ad and joins it to the CRM, so every ad answers for real pipeline dollars, not clicks. Ask for the retrospective and it ranks every campaign by cost per lead and flags the ones whose cheap leads never turn into deals. It builds and places new ads too: created paused, never live, until a human types ACTIVATE.
The tool now runs in the hands of a non-developer on the team. The onboarding doc assumes zero coding ability, because the interface is plain English. That handoff is the point.
“We used to pay a contractor 10% of our ad spend to manage this. Now I ask how the ads are doing and get a straight answer with the numbers behind it. It can even build the next ad while we're talking.”
Built withCustom code on Meta's official Marketing API and the HubSpot API, a database, driven end to end through Claude Code. Read paths and write paths are separate by design; the write path can't put an ad live on its own.
Start where it
pays the most.
Somewhere in your business there's a job that eats a week and shouldn't. That's your first build. Bring it to me and I'll prototype it free. You'll see it working before you spend a dollar.