PharmaMind Pro answers the three questions a prescriber asks before signing: what is this drug and what should I know about it?, what treats this condition, and by which class?, and does this drug carry an interaction or genetic risk I must act on? — without ever computing a dose (that hand-off belongs to DoseOptimize).
Open the app: pharmamindpro.velansai.in (one-time email code; your address is already on the allow-list). Every output below is the tool's actual live result.
Brain-protected: the underlying data sources are concealed throughout — no source is named in the app or here. Reference / decision-support only; the treating clinician retains prescribing authority; model-generated prose is a well-referenced draft requiring clinician review.
1 · Capability catalogue
| Capability | What it does |
|---|---|
| Drug identity & label | Canonical identity, drug class, regulatory-label picture; brand↔generic resolution across several names at once; label-photo OCR |
| Interaction screen (DDI) | Screens a drug against a curated, tiered drug–drug interaction doctrine (Victim ↔ Perpetrator ↔ Rescuer): risk, potential harm, management direction |
| Pharmacogenomic screen (PGx) | Flags drug–gene risks (CYP2D6/2C19/2C9/VKORC1/TPMT/NUDT15/DPYD/SLCO1B1/HLA-B) with phenotype, harm, alternative, CPIC level |
| Disease → drug-class map | A condition's drugs grouped by drug-class stem; treatment-vs-contraindicated separated; treatment targets marked |
| Treatment-drug risk flagging | Marks which of a disease's candidate treatments carry a documented DDI/PGx risk — inline (⚠️) and in a detail panel |
| Contraindication screen | Flags which drugs are contraindicated in named patient conditions |
| AI class monograph | Deterministic, drug-cited, four-section class briefing (no doses; sources concealed) |
| Physician Assistant | Pre-prescription Q&A, Pre-Rx Checklist, Patient Hand-out, multilingual translation |
| Enforced dosing boundary | Declines to dose; defers to DoseOptimize |
★ Flagship — “One statin, two ways to harm”
Starting simvastatin — the interaction and the genotype, caught together
Everyday scenario. A 64-year-old is started on simvastatin. He is also taking clarithromycin, and — unknown to anyone — carries a reduced-function SLCO1B1 transporter variant. Both point at the same catastrophe (rhabdomyolysis) by two mechanisms.
In the tool. Drug Information/Interaction Lookup → simvastatin → FETCH. Below the identity/label output, the 🧬 Interaction & Pharmacogenomic Screen fires:
Feature cases (all with real output)
Case A — The interaction screen from the perpetrator side
Look up clarithromycin. It is flagged as a perpetrator:
| Tier | Clarithromycin × | Potential harm |
|---|---|---|
| 1 | Methadone | Fatal arrhythmia (QT + accumulation) |
| 1 | Colchicine | Multi-organ failure, death |
| 1 | Simvastatin / Lovastatin | Rhabdomyolysis, acute renal failure |
| 1 | Glyburide | Life-threatening hypoglycaemia |
One lookup shows both what harms this drug and what this drug harms — the perpetrator view is where avoidable macrolide catastrophes hide.
Case B — Pharmacogenomic screen: the genotype landmines
| Drug | Gene (phenotype) | Potential harm | CPIC |
|---|---|---|---|
| Abacavir | HLA-B*57:01 (carrier) | Fatal hypersensitivity on rechallenge | A |
| Allopurinol | HLA-B*58:01 (carrier) | SJS/TEN 15–35% mortality; DRESS | A |
| Azathioprine | TPMT / NUDT15 (poor metaboliser) | Life-threatening myelosuppression, sepsis | A |
| Codeine | CYP2D6 (ultra-rapid) | Fatal respiratory depression; neonatal death via breast milk | A |
| Warfarin | CYP2C9 / VKORC1 | Major bleeding, intracranial haemorrhage | A |
The drug–gene pairs where a single unscreened prescription can kill — flagged, with the alternative named, the moment the drug is looked up.
Case C — Disease map + treatment-drug risk flagging
Disease Lookup → Hypertension → 163 drugs across 29 classes, 136 direct treatments (🎯). The treatment table marks risky candidates inline; 4 of 136 are flagged: verapamil, metoprolol, furosemide, hydralazine — each with its interaction/PGx detail (e.g. verapamil as a CYP3A4/P-gp perpetrator; metoprolol CYP2D6).
Case D — Contraindication screen (the topical β-blocker trap)
Disease Lookup → Glaucoma → 🔬 Advanced: Check Drug Contraindications → Asthma. Flags the drugs to avoid: timolol, levobunolol, carteolol, metipranolol (topical β-blockers → bronchospasm) and physostigmine — while prostaglandin analogues, carbonic-anhydrase inhibitors, α-agonists and rho-kinase inhibitors remain 🎯 safe alternatives.
Case E — Class monograph
In the Hypertension result, open the ACE-inhibitor (pril) stem card → 🧠 Generate class monograph → a deterministic four-section briefing (Therapeutic Applications, Mechanism, Physiological Effects, Class & Prescribing), every statement citing specific drugs (benazepril, lisinopril, ramipril, …), no doses, no source names.
Case F — Know the molecule
Drug Lookup → Mounjaro, Zepbound, tirzepatide → all three resolve to one canonical molecule with class and label picture, side by side.
Case G — Physician Assistant + multilingual hand-out
Physician Assistant → metformin → Pre-Rx Checklist; then Patient Hand-out with the translation selector on Tamil (Hindi/Telugu/Malayalam/Kannada/Spanish/Arabic also available). Clinician checklist → patient hand-out → in the patient's language, in three clicks.
Case H — The boundary, demonstrated
Ask the assistant “What dose of vancomycin should I give?” → it declines: dosing is out of scope, defers to DoseOptimize, then continues with non-dosing information only. The boundary is enforced, not a disclaimer.
Design & safety notes (brain-protected)
- Determinism. LLM output runs at temperature 0 with a fixed seed and a cache; an identical question returns byte-identical output. Regenerate is the only path to a fresh generation, which it then re-caches.
- Boundary. The assistant is bypass-proof on dosing — no dose / range / strength / frequency, even from labelling, even if asked directly; it defers to DoseOptimize.
- Brain concealment. No data-source names appear anywhere in the UI or output; live retrieval is routed through a private proxy so the app references only its own host.
- Curated-doctrine scope. The interaction/PGx screen flags what is in the VELANS DDI/PGx knowledge base (a tiered, CPIC-aligned doctrine); a clean result is worded “…in the VELANS doctrine”, not “no interactions exist anywhere”.