Robert Nealan
closin agents: Pipeline Inspection, Forecast Prep, Quote Prep and other agents with their status and last run, beside a feed of recent runs and the deals each flagged
Agents and their latest runs
closin deal inspection grid: every open deal with a health score, owner, stage, status and amount, at-risk deals flagged
Every open deal scored and flagged when it slips
Coffee, closin’s revenue agent, answering “Show me the deals at risk” with four deals, the shared signals and the deals behind them
Coffee answering from the CRM, with the deals behind each answer
closin home: the action feed beside the forecast snapshot, closed-won totals and a deal hygiene gauge
Home: the forecast at a glance
closin marketing site hero: Know why the forecast is soft before the board asks, beside a diagram of acquisition and expansion signals
The new marketing site
That’s all 5 screensKeep going to loop back

Case study · 2026 · Seed-stage revenue operations

closin.ai

closin syncs a company's Salesforce or HubSpot data and puts AI to work on it. I designed most of the product and built its frontend, its design system and the agents themselves.

Role
Senior Design Engineer
When
Mar 2026 – now
Owned
Product design, frontend, design system, AI agents

01

The problem

Forecast prep took someone 4–8 hours a week, and reps on bigger teams routinely missed deals that were slipping. The data was all in the CRM. Nobody had time to read all of it.

02

What we built

01

Four AI agents a sales team opens every morning

Pipeline Inspection, Forecast Prep, CRM Health and Quote Prep, built in two 2-week sprints on Claude (Amazon Bedrock and the Anthropic API) and Gemini on Vertex AI. A sales team runs its pipeline reviews and forecast meetings on them.

02

An editable workflow engine

Scheduling, sequenced tasks, run transcripts and Slack or email delivery underneath every agent. Reps approve, reject or edit each CRM change an agent proposes.

03

Forecasting on a grid that keeps up

Snapshots, scenarios and accuracy tracking, on a grid I rearchitected in under a day to load 200,000+ rows in under a second, with event-based snapshots to view the data at any point in time.

04

An assistant that shows its sources

Coffee, the in-app AI assistant, redesigned so every answer deep-links to the deals and accounts it cites and asks a clarifying question when a request is ambiguous.

05

One design system for every surface

crema: Tailwind and Radix tokens shared by the web app, mobile, email and the assistant, built to WCAG AA.

03

Architecture

Frontend
Next.jsTailwindRadixAG Grid
Backend
tRPCMCP serverAWS Lambda + SQSAWS CDK
Agents
Workflow engineClaude (Bedrock, Anthropic API)Gemini (Vertex AI)
Data
Postgres with row-level security
Services
SalesforceHubSpotSlack
Testing
Playwright

Next case study

Kismet Health

Rebuilding a therapy room for kids