A live product, a production onboarding agent, and a method for modernizing legacy processes. See what changed—and how it works.
Selected projects
Live product
01 / 03
LEADDEMON
Find the companies your CRM is missing.
Built for Stella Software, LeadDemon searches a target market, validates real company websites, compares the results with HubSpot, and prepares likely new accounts for human review.
ACTUAL PRODUCTLeadDemon is live at leaddemon.io—not a concept or a prototype.
In production
02 / 03
STELLA SOFTWARE
Stella onboarding. From days to minutes.
New-client setup meant gathering files, interpreting instructions, and repeating changes across the same app. Winklytics built a Codex-powered agent that helps the team configure company settings, load inventory, and set up taxes and billing in Stella’s existing software.
Back-office users
2
Client migrations / month
~12
From kickoff to live
~1 month
Timing and usage reported by Winklytics’ delivery lead. Results vary with client data and required human steps; this is not a guaranteed onboarding time.
WORKFLOW ILLUSTRATIONThe team brings the client files. Codex and purpose-built tools do the repeatable setup. People handle the exceptions.View full image ↗
Learn the real process.
We walked through the work in the browser, mapped the inputs and decisions, and captured them in reusable agent instructions.
Give the agent better tools.
We built a command-line toolkit: a direct way for Codex to work with Stella’s existing app, so repeatable steps don’t depend on clicking through screens.
Put the team in control.
Staff open a Codex project for each client and add the source files. They run migrations several times a week, handle payments and accounting connections, and usually resolve routine fixes themselves. We support unresolved issues.
Migration method
03 / 03
LEGACY PROCESS MIGRATION
Move trusted logic out of legacy tools.
A practical method for legacy process work: capture trusted inputs, outputs, and business rules; map the data; design a new data model around the process and business needs; then rebuild in Databricks and compare old and new results before handoff.