SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Innovaccer is building an agentic AI platform for healthcare, committing $250M over three years. This role is a senior AI engineer on the core model systems team, working on problems that general-purpose models cannot solve: reasoning over patient records spanning multiple systems and decades, evaluating prior authorizations against dynamic payer policies, and extracting diagnosis codes with clinical accuracy requirements.
You will own one or two deep technical areas while staying conversant across the full stack:
**Model Systems & Product Integration**: Design and own the model portfolio behind product surfaces like Flow (revenue cycle platform). This includes fine-tuning across the full size range—1B to 8B models for latency-sensitive tasks, 8B to 70B for harder reasoning, and 100B+ for frontier capability. You compose these into systems using routing, cascades, retrieval, structured decoding, and verifier models. You own the production lifecycle: versioning, shadow deployment, staged rollout, drift detection, retraining triggers, and regression testing. You hold product-level accuracy bars, not just benchmark scores.
**Post-Training at Scale**: Domain adaptation of open-weight models to clinical, claims, and payer-policy data. This includes supervised fine-tuning, preference optimization (DPO, GRPO), RL with verifiable rewards, continued pretraining, parameter-efficient methods, distillation, and reward modeling. You run large distributed training jobs on hundreds of GPUs using FSDP, DeepSpeed, tensor/pipeline/sequence parallelism, and diagnose production issues in multi-node training.
**Data Pipeline & Quality**: Transform raw healthcare data (HL7 feeds, FHIR bundles, claims files, PDFs, scanned faxes, free-text notes) into training-ready corpora. Design data mixtures, run ablations, generate synthetic data with quality controls, coordinate annotation with clinical experts, and ensure proper decontamination against eval sets.
**Agents & Reasoning**: Build multi-step agents that complete real operational work—submitting authorizations, closing care gaps, resolving denials. Work on tool use, planning, memory, and failure recovery.
You read papers on Monday and have prototypes running by Friday. You also take it to production; there is no research group throwing work over a wall. This is a hands-on role combining research rigor with production discipline.
