Oyster
Recommendations & classificationI worked on skin analysis and recommendations, moved vector search into PostgreSQL, and built the catalog classifier end to end.
Explore the recommendation system03 / ABOUT MEBack to the work
Qudus
Abolade.
I follow the problem
across the stack.
My work connects model development with product engineering. At Oyster, a skincare-recommendation product, I built the catalog classifier end to end and moved product search into PostgreSQL with pgvector.
Serious about the work.
Not necessarily everything else.
Off the clock
01 / 04Formula 1
There’s always a tenth to find.
The precision. The decisions. All that effort behind a very small number.
A perfectly reasonable amount of interest.
Where the work lives.
Different products. Different things that have to hold together.
CalAssist
Calendars & meeting assistanceI built backend integrations for calendar recovery, subscription seat checks, and meeting bots. The work includes tests for rejected billing changes and the bot configuration sent to the provider.
Follow the integration decisionsCuracel
AuntyPelz AI · HR workflowsA Slack assistant prototype connecting policy answers with onboarding checklists, quizzes, and saved progress.
Open the AuntyPelz project noteHow I approach
the work.
Three decisions from Oyster that explain the approach.
JSON SCORING → PGVECTOR
When product search became slow and memory-limited, the useful change was to move vector ranking into PostgreSQL. That made search faster and gave it room to consider more candidates. Relevance still needed its own decisions.
Follow the migration into the productTHE NEXT CONVERSATIONWhat are you
What are you
trying to make work?
I’m interested in AI engineering and forward-deployed roles, working closely with the people using what we build. Tell me about the team, the role, or the problem you want to solve.