Each role starts with the situation, then a task, action, and result you can scan. A link appears only when a published deep dive supports the claim.
Oct 2023–present · Staff Room AI
Founder & consultant · Staff Room AI
Discovery & solution designFirst-of-kind architectureIntegrations & AI configurationDeployment leadershipAdoption & handoffReusable delivery assets
Small operators needed AI systems that matched how work actually moved, and they did not have an in-house team to build them. Each engagement had to end with operators who could keep the system running.
Task. Own discovery through training, and leave a system the client can run.
Action. Mapped the work, built Notion and automation systems, and trained operators at handoff.
Result. Ingestion capacity +99.8% and cost −86%; operators took over the systems.
Aug 2018–Dec 2020 · IBM
Cognitive software and data engineer · IBM
Discovery & solution designIntegrations & AI configurationDeployment leadershipAdoption & handoffReusable delivery assets
Support engineers lost hours preparing client logs before they could analyze a case. Past resolutions were just as hard to find, buried outside the channel where the work already happened.
Task. Surface past resolutions and usable logs inside existing support tools.
Action. Ran discovery with support teams, then shipped a Slack case-matcher and a log pipeline.
Result. Common problems resolved in one Slack response; log-analysis effort fell by an order of magnitude.
May–Aug 2021 · Valhalla Healthcare
Machine learning engineer (intern) · Valhalla Healthcare
Discovery & solution designFirst-of-kind architectureIntegrations & AI configurationReusable delivery assets
Clinicians recorded care in handwritten notes that never became a record the rest of the work could use.
Task. Turn handwritten clinical notes into a data model other workflows could use.
Action. Researched the pipeline, trained a Keras and TensorFlow model, and packaged it in Docker.
Result. Handwritten notes entered an extensible data model at F1 80.03.
May–Jul 2022 · Wiz Music · Barcelona
Community development intern · Wiz Music · Barcelona
Discovery & solution designField-to-product feedback
A music startup read its Discord member count as growth, but the join sources did not support that story. Artists, the reason people would stay, were missing from the server.
Task. Show what the membership data actually said, then test one fix.
Action. Segmented joins by source, raised the inorganic pattern, and shipped an artist AMA.
Result. The team corrected its baseline, and the AMA shipped as a paid artist perk.