01 / 13
RQ
RiskQ360
Risk · Quality · Intelligence
AI Chart Intelligence Platform
RiskQ360
See the whole chart.
Capture every payable HCC.
Evidence-grounded PDF-native Human-in-the-loop
The problem

The evidence is in the chart —
but no one can read every page and prove every code.

Coders are accurate but slow and can't review every chart. Single-model AI is fast but hallucinates codes, misses specificity, and produces nothing an auditor can trace. And most tools need an EHR integration before they see a single page.

40+
pages in a typical chart — more than a coder can review at volume
7,903
ICD-10-CM codes to pick the right specificity from
RADV
audit risk on both under- and over-coding
0
EHR integrations RiskQ360 needs to start
The solution

One loop, from a messy PDF to a payable, defensible code set.

📄

Read

PDF intake with quality scoring and GPT-4o Vision OCR fallback — any source, no integration.

🧬

Predict

BioClinicalBERT + TF-IDF surface candidate HCCs and precise ICD-10s from the full text.

🛡️

Verify

NegEx gating + GPT-4o MEAT check confirms support and removes negated mentions.

💲

Score

V28 HCC mapping, hierarchy, and RAF — plus HEDIS gaps and suspicion signals.

Confirm

Coder Accepts / Rejects / Modifies; every action cited and audit-logged.

How it works · Dual ML + LLM pipeline

No single model guesses the code.
Two engines find it; verification proves it.

ML · BioClinicalBERT

Reads the whole chart

A multi-label HCC predictor works on raw text and catches risk patterns the LLM misses across 40 pages.

  • Predicts HCC categories, not just lines
  • Independent of LLM extraction
  • Trained on CMS-HCC V28
Retrieval · TF-IDF + ICD

Lands the right specificity

Searches the 7,903-code ICD-10-CM catalog, constrained to the V28-mapped subset for each predicted HCC.

  • Matches clinical language to code text
  • Finds E11.65, not E11.9
  • Merged & de-duped with the LLM route
Verify · NegEx + GPT-4o

Proves it with MEAT

Deterministic negation gating, then GPT-4o checks clinical support and Monitor / Evaluate / Assess / Treat.

  • Six-status negation model
  • MEAT evidence per code
  • Confidence score on every code
Grounded code, every time: "E11.65 → HCC 18, RAF 0.302 — supported on page 3: 'type 2 diabetes with hyperglycemia, A1C 8.2%, continue metformin, added glipizide.' MEAT: monitored, assessed, treated. Polarity: active."
ICD-10HCCRAFpage spanMEATpolarity
The processing pipeline

Eight stages. One audit-defensible code set.

01

PDF ingestion

Per-page quality scoring with GPT-4o Vision OCR fallback for scanned pages.

02

LLM extraction ×5

Demographics, clinical sentences, risk dx, HEDIS evidence, encounters — in parallel.

03

ML HCC prediction

BioClinicalBERT predicts HCC categories from the full chart text.

04

ICD-10 retrieval

TF-IDF retrieves precise ICD-10s for each predicted HCC.

05

Negation gating

ConText/NegEx assigns one of six polarities and filters non-payable mentions.

06

MEAT verification

GPT-4o verifies support and extracts Monitor / Evaluate / Assess / Treat.

07

HCC mapping & RAF

V28 mapping, hierarchy suppression, and RAF per encounter & member-year.

08

Output & review

Payable HCC Pack, HEDIS Pack & Audit Pack for one-click confirm.

Inputs & integrations

Reads the charts you already have, where they live.

Chart sources
PDF chartsScanned docsEHR exportsCCDA / CCDDirect upload
Storage
Network foldersAzure BlobAmazon S3SharePointREST API
LLM providers
OpenAI GPT-4oAzure OpenAIGoogle Gemini
Reference data
CMS-HCC V28ICD-10-CM catalogHEDIS specs
ML engine
BioClinicalBERTTF-IDF retrievalConText / NegEx
Persistence
PostgreSQL 16DockerOn-prem / cloud
Frontend
React 19Mantine 8TypeScript
Export
RADV packagesCSV / JSONAudit packs
Domains & lines of business
Medicare AdvantageMedicaid MCOACA / MarketplaceHCC / RAFHEDIS qualityRADV auditSuspicion signalsRevenue intelligence
PDF-native by design — deploys inside the customer environment, on-prem, cloud, or folder-based.
The knowledge-worker cockpit

Not a chat window — a coder's working surface.

● RAF 1.847 Payable HCC Pack

Member · 4 payable HCCs · 1 suppressed by hierarchy · captured-vs-potential tracked

🧬 HCC + ICD

Payable HCCs with supported ICD-10s, RAF weights and hierarchy explanations.

✅ MEAT evidence

Monitor / Evaluate / Assess / Treat per code, with the supporting quote.

📄 PDF viewer + search

Page-level evidence spans traced to source; full-text search across the chart.

🎯 Confidence

ML, LLM and combined confidence per code; negation polarity shown.

🩺 Suspicion signals

Undercoded conditions surfaced from labs, vitals, meds and diagnoses.

📊 HEDIS gaps

Eligibility, numerator and open care gaps across 100+ measures.

🔁 Review workflow

Accept / Reject / Add / Delete / Modify — every action timestamped.

📋 Audit pack export  ·  📈 RAF & coder analytics

RADV-ready package per member, population RAF dashboards, and coder performance over time.

Live demo · Synthetic chart

A 28-page chart with uncontrolled diabetes.

Watch RiskQ360 work the chart end-to-end — from PDF to a coder-confirmed, cited HCC 18.

1

Ingest & OCR

28 pages scored; 3 scanned pages routed to GPT-4o Vision OCR.

2

Extract

Five pipelines pull demographics, sentences, diagnoses, HEDIS evidence & encounters.

3

Predict HCCs

BioClinicalBERT flags diabetes-with-complications from the full text.

4

Retrieve ICD

TF-IDF returns E11.65 over the generic E11.9.

5

Gate negation

NegEx keeps the active mention; drops a "family history of diabetes" line.

6

Verify MEAT

GPT-4o confirms support: A1C monitored, metformin + glipizide treated.

7

Map & score

E11.65 → HCC 18, RAF 0.302; HCC 19 suppressed by hierarchy.

8

Coder confirms

One-click Accept with the page-3 evidence in view; logged for audit.

Built for risk & quality

When a code can't be proven,
revenue and compliance both suffer.

RiskQ360 serves Medicare Advantage, Medicaid MCOs and ACA plans — for coders, compliance and audit teams, quality analysts, and revenue leaders working capitated and value-based contracts.

🔒 HIPAA / PHI-aware · deploys inside your environment · PHI never leaves
  • Risk adjustment (HCC / RAF)

    Prospective & retrospective HCC capture with V28 hierarchy and RAF scoring.

  • Compliance & RADV audit

    Page-level citations and MEAT make every code defensible and export-ready.

  • Quality & gap closure

    Chart-based HEDIS across 100+ measures — COL, CCS, BCS, CBP, TRC and more.

  • Clinical suspicion signals

    Undercoded conditions surfaced from labs, vitals, meds and diagnoses.

  • Revenue intelligence

    Captured-vs-potential RAF, ranked by dollar impact, with recapture tracking.

Adoption maturity ladder

Earn trust one rung at a time.

A coder is always in control of what ships — automation expands only as confidence is proven.

RUNG A

AI-assisted review

RiskQ360 pre-codes and cites; coders confirm every code. Pure decision support.

RUNG B

Confirm-at-scale

High-confidence codes batched for fast Accept; low-confidence flagged for deeper review.

RUNG C

Population intelligence

RAF dashboards, suspicion signals & recapture run across the whole membership.

RUNG D

Closed-loop quality

Gap closure & audit packs feed back into plan operations.Human-confirmed

Outcomes · target metrics

Measurable impact, instrumented from day one.

10×
Faster reviews
Pre-coded, cited charts turn blank-page coding into confirm-and-correct.
HCC recall
The dual ML + LLM pipeline catches supported codes a single model misses.
ICD specificity
TF-IDF retrieval lands the correct sub-code instead of the unspecified default.
RAF capture
Captured-vs-potential closed across prospective and retrospective review.
100%
Audit-defensible
Every code carries page-level evidence and a MEAT breakdown.
0
Unsupported codes shipped
Negation gating & MEAT verification keep over-coding out.
Security & governance

Built so you can trust it in production.

Human-in-the-loop

AI pre-codes; a certified coder Accepts / Rejects / Modifies. No code submitted without a human decision.

Evidence citations

No ungrounded codes. Every ICD, HCC and HEDIS item links to the exact page and span that supports it.

Full audit trail

Every action — who, what, when, before/after — recorded for RADV response and internal review.

HIPAA / PHI handling

Deploys inside your environment so PHI never leaves; sensitive fields access-controlled and governed.

MEAT & six-status negation

Deterministic clinical logic ensures only active, MEAT-supported conditions are coded payable.

Configurable governance

Feature flags, thresholds and prompt templates controlled per plan — and audit-logged on every change.

RQ

RiskQ360 —
read every chart. prove every code.

It reads charts straight from PDFs, codes at the right specificity with a dual ML + LLM pipeline, proves every code with MEAT and page-level evidence, and keeps a coder in control — so you capture what you're owed and defend it on audit.

Let's code a chart together →