BUILDING
Prior authorization & claims, automated
50 states. Millions of possible conditions and treatments. Thousands of payers with changing medical necessity and reimbursement policies. Imperfect and heterogeneous clinical data and sources. US healthcare prior authorization and billing is complex. Foresight encodes rules and policies as logic, with confidence scores and human fallback.
Eligibility
270 / 271
Prior auth
rules + AI
Claims
scrub · submit
Payment
835 / ERA
Weight management · telehealth follow-up
CPT 97803 · ICD-10 E66.01 · 837P
Three sources reconciled. The 270/271 came back covered; the plan's own benefit grid carves medical nutrition therapy out to a third-party administrator. A single check would have addressed this claim to the wrong payer and lost the timely-filing window arguing about it.
Re-assessment documented at 30 minutes, which is what 97803 bills. Rendering taxonomy 1041C0700X is accepted by this plan for MNT and the supervising clinician is on file, so the claim goes out under the group's contracted rate rather than out-of-network.
This plan wants BMI ≥ 30, or ≥ 27 with a documented comorbidity, plus a three-to-six month lifestyle trial before it will pay for continued therapy. The chart carries BMI 31.4 and a T2DM diagnosis. The trial is in the record and not on the claim — which is a CARC 197 denial the moment it leaves.
FixTrial start pulled from the 14 January visit note, attached as documentation, and the claim held for a human rather than submitted. Rework on a denial of this type runs about $48.
Held · 1 fix attached, queued for a human, highest-dollar first
Real-time RAG across 1m+ complex PE documents
PE origination and diligence means reading vast and complex document sets with zero margin for hallucinations. Built for mid-cap funds holding $60bn AUM, Krew.build extracts structured metadata from every document and uses techniques such as deterministic filtering, fast query expansion, hypothetical answers (HyDE), and ranked retrieval and smart merging to generate cited answers in whatever format a team needs (reports, slides, fact-checked docs). We launched a deep research product a year before OpenAI.
Ingest
OCR · layout
Tag
metadata + entities
Index
embed · pgvector
Expand
HyDE + rewrites
Rerank
Cohere · recency
Answer
cited
Real-time form & effort scoring from one camera
Measuring form and effort during exercise normally requires expensive hardware or thin data from wearables. Krew.live scores movement quality and effort-related biometrics from any camera (e.g., phones, laptops) to score effort, and flag limbs that drift from the target (showing where they should be). It also assembles highlight reels from a user's best moments to encourage them to stick to their fitness journey; and provides coaches with tooling to charge for their services.
Too shallow — go deeper
33-point tracking from one camera · smoke marks the position a limb missed