All earlier work ↗
Historical engagement record
About this record
This is an earlier engagement or reference design, preserved as authored. Reported metrics and delivery status have not been independently revalidated for this preview; source dates are not established here. Modeled figures and design targets are not realized results. This record does not substantiate outcomes for the proposed product.
Explore Tensile today ↗
Coding is where revenue-cycle teams feel the squeeze hardest: every chart is manual, every error is money, and “the AI said so” is not an answer an auditor accepts. We built the load-bearing part first: a workspace that lets coders assign, code, review and audit with everything logged. We also built an AI coding proof of concept that links suggested codes to chart text for human review: by linking every suggested code back to the text that justifies it, scoring its own confidence, and running each output past an independent reviewer model before a human signs off. Throughput and accuracy are the goal; explainability is the precondition.
The challenge
Chart coding was manual, slow and costly, with throughput capped at a few dozen charts per coder per day and quality dependent on individual judgment. Leadership wanted AI leverage without surrendering auditability: coders and auditors had to be able to see why every code was assigned. Generic LLM coding was a non-starter, because an unexplained code is unbillable and unauditable.
The approach
A production coder workspace with role-based assign/code/review/audit, real-time collaboration and full audit logging.
An autonomous-coding pipeline: OCR → clinical NLP entity extraction → ICD-10/CPT mapping grounded on clinical ontologies (UMLS/SNOMED) → an LLM coder with confidence scoring, hallucination detection and evidence-linking, so every code points back to the exact chart text.
A three-model pipeline (pre-validation → coding → an independent reviewer model) with a human accept/edit gate on every output.
Artifacts and handoff
A deployed coder workflow application (in production), plus a working autonomous-coding POC running with a launch design partner, including the confidence/evidence-linking layer and the independent-reviewer loop.
Reported measures and design targets
In production: coder workflow app (assign → code → review → audit)
The AI coding proof of concept links suggested codes to supporting chart text for human review.
3-stage pipeline: pre-validation → coding → independent reviewer model, human-gated
POC live: autonomous-coding pipeline running with a launch design partner
Explore the failure mechanisms, the tests they need and the evidence that establishes the customer outcome.
AI reliability, from production to customer outcome.