Support for a person across the entire medication journey — from choosing and checking medicines, to organizing intake, to reaching a doctor.
Significantly raises bleeding risk. Discuss with your doctor before combining.
Every prescription is written by someone who sees only part of the picture. The family doctor doesn't see the cardiologist's changes. The pharmacist sees one prescription at a time. The leaflet is written for lawyers. The patient — taking five medicines from three doctors — is the only one holding the full list, with no tools to check it.
“The data to prevent most of this harm already exists — in leaflets, interaction databases and the patient's own records. It has simply never been assembled around the patient.”
Everything stands on one structured health profile — conditions, medicines, lab results read straight from scanned reports. Three defining capabilities live on top of it and reinforce each other. MedAddWise shortens the path from spotting a medication question to getting qualified help.
One engine — pairwise interactions, whole-regimen assessment, disease and lab risks — can be invoked at every significant step: reviewing the profile, browsing the catalog, reading search results, weighing a basket before purchase, or previewing a new drug with “what if I add…”.
The AI assistant answers instantly, seeing only what the patient chose to disclose. When a human should decide, the case — transcript and consented context included — goes to verified doctors and pharmacists who bid with their own price, protected by escrow.
One tap turns the medication list into a day plan drafted from each drug's official leaflet — dose times, food constraints, spacing between interacting medicines — with every line citing its source. The patient accepts the plan; their phone reminds them.
Nearly every screen below is live in the product: a web application backed by a production API, a populated clinical knowledge graph, and real payment rails. Two aren't screenshots at all — they're the live engine, embedded in this page; type into them. The final two are the layers switching on next.
Patients and clinicians enter 2–10 drugs (and even foods). The engine traverses a clinical knowledge graph to find interaction pathways, then explains each one in plain or clinical language. MedAddWise doesn't stop at flagging an interaction: the system explains why it matters and helps move to the right action — adjust the intake plan, clarify the details, or take it to a doctor. This is not a mockup — the widget beside this text calls the same production API our users rely on.
The check typically takes 5–20 seconds: the engine resolves each medicine, traverses the knowledge graph and writes a grounded explanation. Educational information, not medical advice.
One tap assesses every active medicine against every active condition, recent labs and pregnancy status — and flags what a pairwise checker can't see: disease-modified metabolism and stacked toxicity.
Questions start with the AI assistant — instant, profile-aware, first replies free. When it matters, the case hands off to the marketplace: verified doctors and pharmacists see the question (and consented health context) and bid with their own price. Payment is held in escrow and released on completion.
One tap turns the medication list into an optimal daily schedule: the LLM reads each drug's leaflet and distributes doses across the day — respecting dosing frequency, with-or-without-food requirements, and spacing between interacting drugs. Even a complex regimen — several medicines with different doses, intervals and intake conditions — becomes one coherent, personal plan the patient's phone reminds them about.
A hybrid semantic pipeline (lexical + embedding retrieval, reranking, intent detection) built for Ukrainian medical language. Results carry safety verdicts sourced from official leaflets — for a logged-in user, “safe” means safe for them. Describe a symptom the way a patient would; every stage is reported live by the engine, not animated.
Searches the live medication catalog. Results are informational and ranked by leaflet evidence; logged-in users additionally get personalized safety signals.
Patients upload lab PDFs or photos; our OCR pipeline extracts each analyte, normalizes names and units to LOINC/UCUM, and plots it over time against its reference range — the medical history that usually lives in a drawer.
Access to the health record is granted per person, per data category, with an expiry date — by email, phone or a QR code shown in the clinic. Revocation is instant, and access tied to a consultation ends with it automatically.
The catalog shows live availability across partner pharmacy networks: the patient reserves the product at the nearest pharmacy on the map, or orders delivery. The fulfillment layer — offers, stock, geolocation — is already built into the platform; it switches on with pharmacy-network partnerships.
The next layer we switch on: after an online consult — or instead of one — the patient books an in-person visit with the exact specialist they already trust from the marketplace. The clinic address is provided and verified during specialist onboarding and appears as a pin on the map. The rails underneath — verified specialists, escrow payments, the same geo layer that powers pharmacy availability — are already live.
Escrow-protected like every consultation: funds are held on booking and released after the visit.
No language model on this platform is asked to be the source of truth. Retrieval is deterministic, verdicts are structured, and every generated sentence stands on leaflet text and knowledge-graph evidence — or the system refuses to answer. Below is the graph it all stands on.
Interaction and regimen explanations are written only from retrieved evidence — official leaflet passages and knowledge-graph paths. The output is a structured verdict: safety level, reasoning, recommendations — in patient or clinical wording. With no evidence, the model declines rather than improvises.
grounded · cited · refuses without evidenceThe scheduler reads each drug's official dosing sections and drafts a schema-validated day plan; every slot carries a leaflet-cited rationale. The patient accepts or edits — the AI proposes, the person decides.
structured output · leaflet-citedInstant answers that use only the health categories the patient chose to disclose — undisclosed data never reaches the model. A soft paywall, and a warm handoff that turns the conversation into a specialist consultation.
consent-scoped contextRetrieval and ranking are deterministic — lexical and semantic retrieval with cross-encoder reranking. The LLM only plans ambiguous queries, asks clarifying questions and re-ranks the top results for safety, with red-flag escalation.
deterministic core · LLM at the edgesA dedicated OCR model reads the scan; an LLM extracts the readings into structured, unit-normalized tests that become time series in the profile. Extraction, not invention.
OCR + structured extractionA pairwise lookup table sees "clopidogrel + omeprazole" and shrugs. Our graph walks the biology: a blood-thinner that only works once an enzyme activates it, and a common heartburn drug that shuts that enzyme down. This is one real path, straight from the production graph.
The free safety tools solve a daily problem and create the health profile; the profile makes every consultation and every purchase decision more valuable. The platform takes a share of both — consultations and pharmacy fulfillment. The engine is built — revenue scales with users, not headcount.
20% of every paid consultation. Specialists set their own prices and bid for patients; escrow-style holds (funds captured only on completion) protect both sides and are already live on Ukrainian payment rails.
The catalog shows which partner pharmacies stock a product; patients reserve at the nearest one on the map or order delivery, and the fulfilling pharmacy pays a per-sale commission — the pharmacy-aggregator model already proven at scale in this market. Offers, stock and geolocation infrastructure is built into the platform.
First AI replies are free; the soft paywall converts engaged users, and every AI conversation is a warm lead for a paid human consultation — a built-in upsell path.
Launching in a market of millions of displaced and remote patients where telemedicine is normalized and competition is thin. The clinical engine is dataset-driven and language-portable — each new market is data and localization, not a rebuild.
The same profile and safety engine extend naturally to insurers, clinics and supply-chain integrations — those layers already exist in the platform, switched off until the audience is there.
Under the product is several years of clinical-data engineering that a copycat can't shortcut — and that gets stronger with every user.
A graph of drugs, enzymes, conditions and foods fused from curated biomedical datasets maintained by pharmaceutical research institutions and drug regulators. Finds multi-hop interaction paths no lookup table contains.
graph database · multi-hop traversalA 9-stage pipeline: lexical + dense retrieval, cross-encoder reranking, intent classification and condition detection — engineered for Ukrainian morphology, where global players don't compete.
BM25 + embeddings + rerankEvery official medication leaflet parsed into sections, chunked and embedded — the evidence base the narrator cites, the scheduler doses from, and search ranks by. A data asset that grows with the catalog.
section-level citations · proprietary parsingScanned lab reports become structured, unit-normalized (LOINC/UCUM) time series, with learned aliases improving accuracy with every upload — proprietary structured health data.
OCR pipeline · self-improvingEvery doctor and pharmacist submits a license and credentials reviewed before they can consult. Ratings, escrow and consent-scoped data access make the marketplace safe by construction.
license verification · escrowSearch, health records, consultations, commerce and supply-chain are independent modules on one core — new verticals and new markets switch on without re-architecting.
Swift/Vapor · PostgreSQL · RedisFor investment inquiries, partnerships or a live walkthrough of the product — leave a note and it lands directly in the founder's inbox.
Or write directly: [email protected]