AN Alpesh Nakrani
SolutionsBlogBooksPraiseAbout Work with me ↗
Solutions/ai medical coding assistant

An AI medical coding assistant your coders approve.

The machine reads the note, drafts the codes, and shows its evidence. Your certified coder makes every call, and nothing moves to billing without one.

30 minutes. The price on this page is the price on the call.

Coding Desk illustrative
ENC #2214 · est. office visit awaiting coder
Code set drafted · 99214 + M54.50
CPT E/M guidelines note ¶3 evidence NCCI clean
#2209 · 99213 + J06.9suggested
#2205 · 99204 + E11.9suggested
#2198 · 20610 + M17.11drafting…
99214 · established patient, level 4 trace complete

Note ¶3: “two chronic conditions reviewed, prescription management continued…”

guideline: CPT E/M evidence: 3 note spans coder approval required
Medical coding assistant evaluation dashboard showing 94.6% code accuracy, 96.1% over- and under-coding flag accuracy, all 131 defensibility traces complete, and zero automatic submissions

Golden set: your coded and audited charts · thresholds are contract acceptance criteria · illustrative values

Codes approved · routed to billing
Coder gate: enforced in code No auto-submit · ever
$28.83B
improper Medicare fee-for-service payments in one year, a 6.55% error rate
CMS FY2025 Improper Payments Fact Sheet
$853M+
improper payments on established office visits, 65% from incorrect coding
CMS CERT 2025 data, via AAPC
$65,007
average annual salary of a medical records specialist
AAPC 2025 Salary Survey

Read those numbers together: the leak is measured in billions, coding capacity is priced like a scarce hire, and every miscoded chart went past a human who never had the guideline open at the right paragraph. Throughput is the constraint. Judgment is the asset.

The problem

Coding is the bottleneck between care and cash

Nothing bills until it is coded. Code above what the documentation supports and you invite payer audits and paybacks. Code below it and you write off earned revenue silently, chart after chart. Every RCM firm and coding-heavy practice responds one of three ways:

Route 01 · hire

Hire more coders

A first-year hire averages $45,377, still needs seasoning, and the market is short. Capacity that resigns takes its judgment with it.

doesn’t scale
Route 02 · outsource

Outsource per chart

Staffed coding services absorb volume, and they charge on every chart, forever. The capability never becomes your asset.

fees forever
Route 03 · rent

Rent an autonomous platform

Fathom, CodaMetrix, Nym: serious systems with KLAS awards and enterprise contracts. None publishes a price, and human review is the exception, not the rule.

no public price
Route 04 · own

Own a scoped coding assistant

It drafts every code and submits none of them. Fixed scope, fixed price, eval suite included.

the fourth option →
How it works

A CPT/ICD suggestion pipeline, fixed scope, delivered in seven weeks

  1. 01

    Ingest

    The pipeline reads your chart and note exports (one EHR export format) and assembles encounter context with no manual re-keying.

  2. 02

    Retrieve

    A RAG layer over the current CPT and ICD-10-CM guidelines, NCCI edits, and your payers’ policies for one specialty pulls the rules that govern the chart in front of it.

  3. 03

    Suggest

    The agent drafts the code set with a rationale per code: the note evidence it relied on, the guideline passage that supports it, and a flag wherever documentation supports a higher or lower level.

  4. 04

    The gatehuman judgment

    A certified coder reviews, edits, and approves every code before anything moves to billing. Enforced in code, not in a policy document. The system cannot submit. That constraint is permanent and non-configurable.

Scope is deliberately narrow: one specialty, one EHR export format, that specialty’s core CPT and ICD-10 code families. That is what makes the price fixed and the timeline honest. Expansion is a follow-on, priced the same way, in the open.

The differentiator

The eval suite is the product

Every vendor in this market claims an accuracy number. None hands you the machinery to verify the claim on your own charts. My acceptance criteria are written into the contract as eval scores, and you keep the machinery that produces them.

Check 01built from · your coded + audited charts

A golden dataset

Built from your historically coded charts, including those that survived or failed payer audits. Not synthetic examples.

Check 02catches · over- and under-coding

Code-accuracy regression

Suggestions scored against the golden set, per code family, with over-coding and under-coding tracked separately. Exact-match alone hides the errors that trigger audits.

Check 03fails · any code without a trace

Audit-defensibility tests

Every suggested code must trace to note evidence and current guideline text. A code without a defensible trace fails the suite, whatever the aggregate accuracy says.

Check 04re-runs · every Oct 1 and Jan 1

Update regression

ICD-10 changes every October 1 and CPT every January 1. The suite re-runs after each release, so you find the drift before a payer does.

Check 05blocks · cutover below threshold

A parallel-run gate

The assistant runs alongside your coders on live charts until it clears the thresholds. No cutover before the score clears.

When I hand over, you own the eval harness, the golden dataset, the regression suite, and a runbook. You do not have to trust my accuracy claim, or anyone’s. Re-run the suite: this quarter, next quarter, after every model update. A coding engine without a measurable judgment gate is an audit finding waiting for a date.

The price

The price is $32,500. Here is exactly what it buys.

One price. No tiers, no “starting at,” no quote theater.

One public price
$32,500
one-time · 7 weeks + 30-day stabilization
= half of one certified coder’s average year (AAPC 2025: $65,007), paid once
Schedule a call
  • Discovery workshop: your specialty code map, your top denial-driving codes, where the coder gate sits
  • Chart and note intake pipeline for one EHR export format
  • Guideline RAG: CPT, ICD-10-CM, NCCI edits, and your payer policy set for one specialty
  • Code-suggestion agent with per-code rationale and evidence links
  • Coder review-and-approve workflow (the enforced gate)
  • Golden dataset built from your historically coded and audited charts
  • Eval harness: code accuracy, over/under-coding detection, audit-defensibility traces
  • Guardrails in code: no auto-submission, PHI-handling constraints, confidence-floor routing to your coder
  • Parallel-run period against your current coding process, with published thresholds
  • Handover pack: runbook, eval documentation, your team trained to re-run the suite
What’s not in the number, said plainly: you pay your own LLM API and hosting costs, typically $300–$800/month at billing-firm volume, on your accounts, visible to you. Optional monitoring retainer after stabilization, covering the October and January code-set updates: $2,000/month, cancel anytime. Neither is required to run what you own.
The field

Against the alternatives

This build Autonomous platform Outsourced coding Another coder hire
Price $32,500, once, public Undisclosed; enterprise contracts Per-chart fees, forever $65,007/yr average, plus benefits
Timeline 7 weeks + 30-day stabilization Months of procurement and integration Fast, but never yours Months to recruit in a shortage
Who owns it You: code, evals, golden set The vendor The vendor Capacity that can resign
Proof it works Eval suite you can re-run Their case study, their KLAS badge Sample audits A certification and an interview
Coder control Every code approved by your coder Exception-based review Outside your walls Full, but throughput-capped

For an RCM firm the economics multiply: one owned build, deployed across every practice you code for, with the same eval suite proving it on each client’s charts. I will not promise your accuracy number; that is what the parallel run measures. I will show you the score before you cut over.

Delivery

How the seven weeks run

  1. Week 1

    Discovery

    Map the coding workflow, pull historically coded charts and audit outcomes, fix the code families in scope and the acceptance thresholds in writing.

  2. Weeks 2–3

    Golden dataset + guideline RAG

    Label the historical charts, ingest CPT, ICD-10-CM, NCCI edits, and payer policies, stand up the intake pipeline.

  3. Weeks 4–5

    Suggestion agent + gate

    Per-code rationale with evidence links, the coder review-and-approve workflow, guardrails in code.

  4. Weeks 6–7

    Evals, then parallel run

    Harden the suite against the golden set, then run the assistant alongside your coders on live charts. Cutover only when thresholds clear.

  5. Days 1–30 after

    Stabilization

    I watch the evals, fix drift, tune the confidence floors. Then you own it outright.

FAQ

Straight answers

The questions every coding manager and billing-firm operator asks before booking the call.

Does the AI code and submit charts automatically?

No, and it cannot be configured to. Every code requires certified-coder approval before it moves to billing, and the constraint lives in code. Autonomous submission is how coding tools end up in audit findings; suggestion plus human judgment is how they survive them.

What exactly does the $32,500 include?

Everything in the scope list above: discovery, intake, guideline RAG, suggestion agent with rationale, coder gate, golden dataset, eval harness, parallel run, handover, and 30 days of stabilization. The only costs outside the number are your own API and hosting spend and the optional monitoring retainer.

How do you prove accuracy on our charts, not in a demo?

The golden dataset is built from your historically coded and audited charts, and the contract's acceptance criteria are eval thresholds on that set. Then a parallel run tests the assistant against your coders on live charts before anything cuts over.

Which specialties and EHR formats are in scope?

One specialty and one EHR export format per build; that is what keeps the price fixed. On the call we check whether your specialty mix, note quality, and export path fit. If they don't, I'll tell you, and you'll have lost 30 minutes, not a deposit.

Is this HIPAA compliant?

The architecture is BAA-ready: PHI-handling guardrails in code, minimum-necessary retrieval, and the option to run on infrastructure you control. Compliance obligations and the BAA itself are scoped in discovery and stated plainly, not hand-waved.

What happens when ICD-10 and CPT code sets update?

ICD-10-CM updates every October 1 and CPT every January 1, and that is when un-evaluated coding tools quietly rot. The suite you own includes update regression: refresh the guideline corpus, re-run the tests, read the diff. Do it yourselves from the runbook, or the $2,000/month retainer covers it, cancel anytime.

Why is the price public when every vendor quotes custom?

Because scoped work can be priced, and unscoped work should not be sold. The scope on this page is genuinely fixed, so the price can be too. If discovery reveals your situation needs more, you get a new scope with a number on it before any work starts.

Next step

Suggestion is automation. Submission is judgment.

The autonomous-coding vendors are racing to raise the percentage of charts no human touches. The number that wins audits is different: the percentage of suggested codes your coder can defend, with the evidence attached. That number is measurable, and the machine that produces it is buyable, once.

30 minutes · the price stays $32,500 · if it’s not a fit, I’ll say so

Ask AI about AI Medical Coding Assistant