About the lab
BeamSpark Claims Lab exists because insurers kept buying models that never touched production—or touched it once and frightened adjusters. We are a Korea-based collective of seven practitioners who only take claims-adjacent work. Principles: shadow before swap, document the boring seams, and let SIU voices veto unsafe automation.
Org chart
Engagement spine with delivery pods branching under architecture.
Hana Seong
Engagement Director
Noa Kim
ML Deployment Architect
Daniel Ahn
Claims Domain Consultant
Mira Lee
Client Success Lead
Leo Park
Platform Integration Engineer
Rina Cho
MLOps Engineer
Sora Watanabe
QA Automation Specialist
Field notes (masonry)
“Their intake runway workshop exposed duplicate OCR contracts we were paying for twice. Politically awkward but factually correct.”
“Monitoring lab gave us alert tiers we could staff. First week still noisy, but the RACI page stopped the blame ping-pong.”
“War room week was crisp: pre-reads were short, disagreements logged without theatre. We still disagree on budget, but not on facts.”
“Severity studio pushed us to document leakage tests we had hand-waved before. Launch felt incremental instead of a cliff.”
“Governance kit for copilots was dense. Legal appreciated the logging schema; adjusters wanted more UX mockups next iteration.”
“We wanted someone to say no to a reckless go-live date. They did, with a calendar that finance accepted.”
Principle · Evidence first
Claims metrics beat generic ML leaderboards.
Principle · Adjuster-safe
Overrides stay obvious, never buried three dialogs deep.
Principle · Honest seams
If batch windows block real-time dreams, we say so plainly.
Milestones
- 2019 — First Korea workshop series on claims document automation for a regional TPA.
- 2022 — Formalised MLOps audit template adopted by two national carriers’ risk forums.
- 2025 — BeamSpark Claims Lab incorporated as advisory practice with vendor-neutral ethics charter.