Coding engines built on American charting fail on the vocabulary Indian consultants actually write, pushing coders toward unspecified catch-all codes that are payable but reliably queried. A deterministic engine with a proper Indian vocabulary resolves offline in about a millisecond, costs nothing per call, and cannot invent a code that does not exist.
The vocabulary problem
An Indian consultant writes "sugar +, on OHA" or "k/c/o koch's" or "c/o loose motions × 3 days". These are not sloppy notes; they are a working shorthand every Indian clinician reads instantly and almost no imported engine resolves.
| Written as | Means | Resolves to |
|---|---|---|
| sugar | Diabetes mellitus | E11 family |
| koch’s | Tuberculosis | A15 family |
| loose motions | Diarrhoea | A09 / K52.9 |
| daad | Tinea | B35 |
| kamzori | Weakness, malaise | R53 |
| BP | Hypertension | I10 |
And then the diagnoses themselves. Dengue, typhoid, malaria, tuberculosis and scrub typhus are routine in an Indian OPD. In an engine built around American epidemiology they sit in the long tail, ranked below conditions an Indian clinic may see once a decade.
Why not a language model
It is tempting, and it will appear to work. Three reasons against it for this specific job:
It can invent codes. A generative model asked for an ICD-10 code will always produce something shaped like one. Whether that code exists is a separate question. A deterministic lookup cannot return a code that is not in the table.
It needs connectivity. Clinics in tier-2 and tier-3 cities do not have reliable links, and a coding step that fails when the connection drops is a coding step that gets skipped.
It costs per call. A busy OPD codes hundreds of encounters a day. Per-call pricing on a routine operation is a meter running against your margin.
This is not a compromise. For this particular job, deterministic is strictly better — and it keeps the tool clear of clinical-decision-support framing, which carries its own regulatory weight.
The unspecified-code trap
When a phrase does not resolve, the coder reaches for an unspecified catch-all. It is payable, so nothing appears to go wrong. It is also the category TPAs query most consistently, so the cost arrives later and looks like a documentation problem rather than a coding one.
Worth measuring: what share of your codes are unspecified. A coverage report that surfaces the phrases the engine keeps missing tells you exactly where the next improvement is.
Discharge summaries that hold up
A summary that exists but says nothing gets queried as often as a missing one. Four things separate the two:
- Hospital course composed from the treating doctor’s own dated notes, never a generated narrative. Provenance is what makes it defensible.
- Abnormal results named, normal panels collapsed. A wall of normal values buries the finding that justified the admission.
- Discharge medication limited to what is still running, not everything ever given during the stay.
- Every section traceable to the record behind it.
What compounds
Every practice has its own shorthand on top of the shared one — a particular consultant's abbreviations, a department's habits. An engine that learns those after the second use gets more useful the longer it runs. One retrained centrally every quarter never learns your building at all.
More in this cluster: Coding & documentation — every article we have on it.
Questions we get asked
Is the coding engine AI?
No, and deliberately so. It is deterministic — it runs offline in about a millisecond, costs nothing per call, and cannot produce a code that does not exist.
Why do unspecified codes matter if they are payable?
They are the codes TPAs query most consistently. The cost arrives weeks later as a documentation query rather than at the point of coding.
Does it work without an internet connection?
Yes. The coding step makes no network call, which matters for practices in tier-2 and tier-3 cities where connectivity is unreliable.
Read next
See it against your own settled claims
Twenty claims you have already settled, showing what was disallowed, how much was predictable, and how much the payer never explained. About an hour of your team’s time.