TR EN
Contact

Regulation intelligence for healthcare

Where the substance meets the article.

Madde AI turns healthcare regulation (SUT, the EK-4/D lists, TİTCK notices) into executable, pharmacist-reviewed rules, and checks every prescription, medical report and drug order in seconds. As SaaS, or fully on-premise.

Contact See EczaneRapor (opens in a new tab)
madde.ai · live check Synthetic example
Setting
Pharmacy
Drug
Atorvastatin 40 mg
Diagnosis
E78.0 · Pure hypercholesterolaemia
Finding
LDL 192 mg/dL · 14 Aug 2026
Report
Specialist report · Cardiology

SUT 4.2.28.A Lipid-lowering drugs (abridged)

Statin therapy may be started in patients whose LDL cholesterol, measured within the last 6 months, is 190 mg/dL or higher. The lab value and its date are stated in the report. The drug is prescribed on a report issued by a specialist that is still valid.

  • EK-4/D diagnosis match met
  • LDL threshold and test date met
  • Report within validity period met
  • Dose matches the report met
ELIGIBLE

Every condition is documented in the report. Basis: SUT 4.2.28.A

Setting
Pharmacy
Drug
Adalimumab 40 mg
Diagnosis
M45 · Ankylosing spondylitis
Finding
BASDAI value not entered
Report
Health board report · Rheumatology

SUT 4.2.1.C TNF inhibitors in ankylosing spondylitis (abridged)

Started in patients whose disease stays active despite at least 3 months of NSAIDs, with a BASDAI of 4 or higher on two measurements 4 weeks apart, on a health board report that includes a rheumatology or PM&R specialist. Baseline BASDAI values are stated in the report.

  • Diagnosis matches the article met
  • Specialist health board report met
  • Baseline BASDAI value not found in the report missing
MANUAL REVIEW

The baseline BASDAI value is missing. No guessing: the case goes to the pharmacist for review.

Setting
Hospital · Inpatient order
Drug
Intravenous immunoglobulin (IVIG)
Diagnosis
D69.3 · Immune thrombocytopenia
Dose
1 g/kg/day · 2 days · 68 kg
Report
Health board report · Haematology

SUT · IVIG & barcode reporting Principles of use (abridged)

In immune thrombocytopenia, IVIG is given on a health board report, not exceeding 1 g/kg per day for at most 2 days. For drugs used in inpatient care, barcode (karekod) data is reported to Medula before billing.

  • Report match met
  • Dose ceiling (1 g/kg/day) met
  • Barcode label (küpür) not sent to Medula failed
NOT ELIGIBLE · deduction risk

The barcode label of 1 of 4 vials was never reported. Fixable before the invoice goes out.

Synthetic example; no real patient data. Article texts are abridged.

Madde AI’s engine, at work every day inside EczaneRapor.

pharmacies
4,000+
prescriptions evaluated
2M+
accuracy
98%
fewer false approvals¹
5×

¹ Compared with a general-purpose large language model; blind test on 3,000 unseen cases.

§ 00The word

§
madde¹ · article madde² · substance madde.ai

madde

/mahd·deh/ n. (Turkish, from Arabic mādda)

  1. An article of a law or regulation. e.g. “SUT madde 4.2.28”, SUT article 4.2.28
  2. A substance; the active ingredient of a drug. e.g. “etken madde: atorvastatin”, the active ingredient atorvastatin
  3. madde.ai The AI that connects the two.

In Turkish, one word names both the article of a regulation and the active substance of a drug. Madde AI checks every active substance against every article that governs it.

§ 01Problem

Every drug is bound to an article. The articles keep changing.

In Turkey, every reimbursed drug and procedure is bound to an article of SGK’s Healthcare Implementation Communiqué (SUT). Hundreds of articles, the EK-4/D diagnosis lists and TİTCK notices have to be read together, and the text changes constantly.

  1. 01

    The text never stands still

    SUT articles, annex lists and notices change throughout the year; a check that was right yesterday can cause a deduction today.

    30 article texts changed between December 2025 and September 2026. In a single day of real traffic, 215 distinct SUT paragraphs came into play.

  2. 02

    Manual checking doesn’t scale

    Pharmacists and hospital billing units cannot hold thousands of reports a month against the regulation by hand. Every missed condition comes back as a deduction at month-end.

  3. 03

    Deductions arrive at month-end

    An SGK deduction means lost revenue, rework and month-end stress. Yet most of it stems from gaps that could have been caught before submission.

Deduction breakdown at a partner hospital (TL) 17,115 deduction lines · TL 3.75M
  • Documentation & barcode-label gaps 60%
  • Procedure coding 12%
  • Justification & discharge summary 8%
  • Clinical SUT conditions 8%
  • Other 12%

About 60% of the money was lost to documentation and barcode-label gaps, all preventable before submission.

The hospital is kept anonymous.

§ 02How it works

Rules first, model second.

Madde AI never leaves the decision to a language model’s guess. It compiles the regulation into rules first; models step in only to read the documents and to fill gaps that genuinely need interpretation.

  1. Compile the regulation

    SUT, EK-4/D and TİTCK texts become versioned decision trees, each reviewed and signed off by pharmacists. When the regulation changes, the rule changes with it.

    SUT · EK-4/D · TİTCK

  2. Extract the facts

    Our own fine-tuned language models read the report and the prescription and pull out values, dates and diagnoses. Numbers and dates are verified in code, never guessed.

    values · dates · diagnoses

  3. Decide

    Deterministic rules make the call. Where evidence is missing the verdict is MANUAL REVIEW, never a guess. A model is consulted only where a gap genuinely needs interpretation.

    rule → verdict

  4. Explain and audit

    Every verdict cites the exact article and the evidence behind it. The whole process is kept in an audit trail.

    article · evidence · log

Every result is one of three verdicts

  • ELIGIBLE Every condition is documented.
  • MANUAL REVIEW Evidence is missing; a pharmacist decides.
  • NOT ELIGIBLE A condition is not met; deduction risk.

§ 03In the field

Proven in the field: EczaneRapor

EczaneRapor

Founded with pharmacists

Madde AI’s engine already powers EczaneRapor, used by 4,000+ pharmacies across Turkey.

  • Reads SGK reports straight from Medula
  • Works with every pharmacy automation system: Botanik, TEBEOS and more
  • Set up in minutes
  • Rated 4.9/5 by pharmacists
eczanerapor.com (opens in a new tab)

§ 04Next: hospitals

The same engine, hospital-grade.

In a hospital, Madde AI steps in at two critical points: before a drug reaches the patient, and before the invoice reaches SGK.

Checkpoint 1

Before the hospital pharmacist approves

Each inpatient drug order is checked for dose, report and interactions before the hospital pharmacist signs it off.

Checkpoint 2

Before billing submits to SGK

The invoice file is screened for missing documents, unreported barcode labels and deduction risk before it goes to SGK.

Six checks

  • Dose ceiling

    Maximum daily and per-course doses, adjusted for weight and age.

  • Report match

    Is the order backed by a valid health board or specialist report?

  • Infectious-disease approval

    Specialist sign-off for restricted antibiotics.

  • Drug interactions

    Clinically significant interactions across a patient’s concurrent orders.

  • Barcode-label (küpür) submission

    Has the barcode label of every administered pack been reported to Medula?

  • Non-reimbursable drugs

    Drugs SGK does not reimburse are flagged before billing.

Evidence

  • Read-only integration with the HBYS (Oracle)
  • Validated on a partner hospital’s live data
  • 99.7% of pending drug orders matched to the national drug database by barcode

§ 05Deployment

In the cloud, or inside your own walls.

SaaS

For pharmacies

Ready in minutes

  • Works with every pharmacy automation system
  • Medula integration
  • Secure, encrypted infrastructure
  • Regulation updates arrive automatically
pharmacies use it
4,000+
pharmacist rating
4.9/5

On-premise

For hospitals and institutions

Data never leaves

  • Runs inside the hospital network
  • Our own fine-tuned open-weight models (7B–70B), served locally with vLLM
  • No patient data ever leaves the institution
  • KVKK-ready, with complete audit logs
Hospital network HBYS Madde AI vLLM · 7B–70B Billing unit Internet Patient data stays inside Hospital network HBYS Billing unit Madde AI vLLM · 7B–70B Patient data stays inside Internet

§ 06Company

Kyanalitik Yazılım

We are an engineering team in Teknopark Ankara, working side by side with pharmacists. We build sovereign language models that run on-premise, inside the regulated institutions that use them.

Madde AI is Kyanalitik’s healthcare-regulation product family; EczaneRapor is its product for pharmacies.

  • Teknopark Ankara
  • Built with pharmacists
  • On-premise language models
kyanalitik.com (opens in a new tab)