YC W24
LLM Workflow Engineering at Artisan
Less busywork.More possibility.
First hire at Artisan. Built the early outbound systems that turn prospect research into personalized outreach.
Behind the experience
Explore the flowUseful context comes first.
Prospect information gives personalization something to work with.
I built scraping and data-enrichment workflows to assemble context before asking a language model to write outreach.
Make each message relevant.
Language models turn context into tailored outreach.
I worked on generation and prompt evaluation, and used batching and parallelization to improve workflow throughput.
Meet the existing workflow.
The result connects to the communication tools teams use.
I integrated and maintained services including Apollo, Gmail, Outlook, Warmy, MailGenius, and SendGrid.
An illustrated overview of the work. Select a stage to look closer.
Research
01 / The challenge
What had to
feel simple.
Artisan's Ava connects prospect research, personalized emails, and outbound campaigns. Personalization needs more than a prompt: useful information about each prospect, generation grounded in that context, and a dependable route into the tools people already use.
02 / My contribution
Where I got
my hands dirty.
Building the early product
As Artisan's first hire, built the web-scraping, enrichment, and LLM generation pipelines behind personalized outbound workflows.
Throughput without losing context
Improved large-scale workflow execution through batching and parallelization, reducing processing from hours to minutes.
Connecting the last mile
Integrated the communication and enrichment services needed to support an end-to-end campaign, and maintained those connections as the product grew.
03 / The engineering idea
Personal at the edges. Parallel underneath.
Each prospect needs individual context, but independent work doesn't need to wait in a single queue. Batching and parallel execution create room for personalization without making an entire campaign move one request at a time.
What stays with me
The useful part of an AI workflow is the whole loop: relevant input, well-directed generation, and a result that arrives where people need it.
Stack AI
Execution engines · Product infrastructure