Tanner Velten

Backend engineer on the in-house claims platform that replaced Oscar's COBOL vendor stack. Runs 98.7% first-pass accuracy across roughly $10B in annual medical spend, for a book that grew from 400K to 3M members during my time here. I designed the benefits DSL that moved 70% of rule changes to ops, and the reconciliation harness that gated a four-market migration.

Eight years in Go, Java, and Python on HIPAA-regulated money systems, four of them in Go. Looking for senior, staff-level platform work in claims or payments.

Senior Software Engineer II, Tech Lead, Claims at Oscar Health (2025-08 – Present)

• Took ownership of the pricing and benefits contract for a platform running 98.7% first-pass accuracy with most claims paid inside 48 hours, serving 3M members and ~$10B in annual medical spend.
• Made preview the required path for all 270/271 eligibility and cost estimates. Same pipeline as a live 837 with payments disabled, with weekly CI contract tests blocking drift.
• Designed the agent layer so its tools call the pricing and benefits services directly rather than reimplementing them, which makes a quoted claim and a paid claim structurally incapable of disagreeing.
• Gated agent responses behind offline evals against the golden-claim corpus and a human-in-the-loop hold on denials, on the argument that an agent allowed to deny without review is a compliance problem before it is a product one.
• Set the review bar for the pipeline: no change ships without a golden-claim diff in CI. Still write DSL runtime, pricer, and reconciliation code, and carry on-call escalation.

Senior Software Engineer, Tech Lead, Claims at Oscar Health (2023-09 – 2025-08)

• Cut benefits time-to-production from 11 days to under 24 hours and moved 70% of benefits changes off engineering, by designing and writing a DSL that lets ops author deductible, coinsurance, copay, OOP max, and accumulator rules with no deploy.
• Drove new-versus-legacy mismatch from 6.1% to 0.4% across a year-long parallel run, then cut over four markets with zero missed regulatory payment deadlines.
• Argued for line-level reconciliation as a CI gate rather than a pre-cutover audit, which meant every pipeline change was checked against legacy output continuously instead of once per market.
• Kept remittance whole through the Feb 2024 Change Healthcare outage by adding idempotent hold-and-flush to the 835 rails. Files queued through the outage and drained clean with no double payments.
• Led the Pricing and Benefits pod, six engineers: design reviews, on-call rotation, two mid-levels onto first line on-call. Go and Java on AWS/EKS, Kafka ingest, stateless synchronous pipeline over a temporal claim store, EDI 837 in and 835/EOP/EOB out.

Software Engineer, Claims at Oscar Health (2022-05 – 2023-09)

• Cut p99 pricing latency from 420ms to 90ms by rebuilding pricing as a stateless Go service with Java adapters over the legacy fee-schedule and contract stores.
• Made that service the single allowed-amount path for both live claims and member estimates, taking the vendor black box off the hot path for 837P and 837I.
• Built replay against the temporal claim store so any raw 837 could be re-adjudicated after a config or code change. It is now how the team validates every pipeline change.
• Accepted higher per-claim compute in the stateless synchronous design, trading throughput headroom for the ability to replay and reason about any claim at any time.

Software Engineer, Virtual Care at Oscar Health (2020-11 – 2022-05)

• Shipped $0 Virtual Primary Care across 10 markets for plan year 2021: longitudinal PCP, labs, imaging, referrals, tier-1 drugs, at-home vitals.
• Implemented the real-time cost-share waiver that zeroed member cost at order time when an Oscar PCP ordered a covered service.
• Built Python and TypeScript services and React surfaces for the Oscar Medical Group visit flow through the COVID telehealth surge, under HIPAA authz, audit logging, and encryption.

Software Engineer at Caspar.AI (BrainofT) (2018-08 – 2020-11)

• Owned a household event pipeline end to end at an 11-to-50-person senior living smart home company: device ingest, alerting, CI, and production on-call. Stored derived events only, never raw device media.

B.S. in Computer Science – UCLA (2014-09 – 2018-06)