Clinical decision support · RAG over FDA labeling

Drug safety answers you can trust — grounded in the label, never guessed.

Pill IQ gives pharmacists and physicians instant, cited answers on allergies, interactions, and alternatives. Every response is grounded in FDA drug labeling and validated — when the label doesn't say, neither do we.

Request a demo See it in action

✓ OpenFDA + DailyMed ✓ Every answer cited ✓ No hallucinations

Pill IQ · Interactions

warfarin and ibuprofen

✕ CONFLICT

Severity: major

Warfarin labeling lists ibuprofen (an NSAID) as increasing bleeding risk when used concomitantly; ibuprofen warns that anticoagulants raise the chance of severe stomach bleeding.

Sources

[1] dailymed.nlm.nih.gov · Warfarin Sodium

[2] dailymed.nlm.nih.gov · Ibuprofen

What it does

One tool, three high-stakes questions — answered cleanly.

Each mode returns a structured, glanceable verdict your team can read across the counter in under a second. No paragraphs to parse.

Allergy Check

“Can a patient allergic to X take Y?” → a clear SAFE / NOT SAFE verdict with the cross-reactivity reason.

Interaction Check

Check two or more drugs at once → CONFLICT / NO CONFLICT, severity scale, and every interacting pair spelled out.

Find Alternatives

“Can't take X — what else treats this condition?” → a ranked list of alternatives, each with a one-line rationale and its own label source.

Source-linked evidence

Click any citation to open the exact label passages it relied on — relevant terms highlighted, the live DailyMed page alongside.

Typo-safe by design

Look-alike names (citalopram vs. escitalopram) trigger a quick “did you mean?” picker — so a misspelling never becomes the wrong drug.

Ask anything, grounded

A general mode answers free-form medication questions straight from the labels — and politely declines anything outside that scope.

How it works

Retrieval-grounded, schema-validated, cited.

The model never free-associates. It answers only from retrieved label text, and every response is validated against a fixed schema before it reaches you.

Step 1

Ask in plain language

Type a question or call an agent with /. Names are normalized to RxNorm so brands, generics, and typos resolve to one drug.

Step 2

Retrieve the right label text

A vector search over FDA labeling (OpenFDA + DailyMed) pulls the exact contraindication, interaction, and indication passages — pre-filtered to the relevant drugs.

Step 3

Answer, validated & cited

Claude returns a strict JSON verdict, checked server-side. No valid grounding? You get INSUFFICIENT DATA — never a guess.

Built for safety

Accuracy is the product. A wrong answer is worse than no answer.

Pill IQ is engineered so the bias always favors caution. Every layer — retrieval, prompting, validation — is designed to refuse rather than improvise.

✓Grounded only. The model answers from retrieved label context and nothing else — no outside knowledge.
✓Cited, always. No source means no verdict. Every answer carries the labels it stands on.
✓Versioned evidence. Each fact records its source and ingestion date — you can prove what the tool knew, and when.

? INSUFFICIENT DATA

When the label doesn't contain enough to answer, Pill IQ says so plainly instead of filling the gap. For a clinical tool, “I don't know” is a feature.

Pill IQ supports — it does not replace — professional clinical judgment. v1 reflects FDA labeling and is intended for licensed professionals.

For builders

Bring trustworthy drug intelligence to your agents.

Pill IQ ships an MCP server, so Claude Desktop, Claude Code, or any agent you build can call the same grounded, cited pipeline as four clean tools.

allergy_check interaction_check find_alternatives ask_medication

terminal

# add Pill IQ to any MCP-capable agent
claude mcp add pill-iq \
  --env PILLIQ_API_URL=http://localhost:8000 \
  -- python mcp/server.py

# → tools available to your agent:
interaction_check({ medications: ["warfarin","ibuprofen"] })
→ { verdict: "CONFLICT", severity: "major", sources: [...] }