[synthesis.] is an internal regulatory intelligence workspace: a local catalog plus live APIs for labels, trials, and literature. Research skills run in Cursor via ToolUniverse — not inside the web UI. Not medical or legal advice.
These feeds populate Postgres (`DrugProduct`, sponsors, canonical rollups). Coverage depends on what you run locally (`npm run ingest`, DailyMed jobs, etc.).
Structured drug labeling (SPL) from NIH DailyMed — primary path for rich indication and packaging text in the local catalog when that pipeline is enabled.
BLA / biologic product metadata (CSV export or normalized extracts) used to seed and reconcile biologic rows and canonical keys.
Optional Purple Book–style or sponsor-normalized CSV ingests for controlled corpora and demos.
Ask can query these at request time (networked). Authenticated keys improve throughput where vendors allow it.
Public FDA drug labeling API (SPL-backed JSON). Used for live label discovery, infusion/rare-disease baselines, and optional enrichment during ingest — not the only regulatory surface in the product.
Application-centric regulatory metadata (submissions, marketed product rows) surfaced in diligence panels where configured.
Parallel retrieval for trials and publications; complements catalog and label rows.
Trial registry links and metadata for landscape-style Ask turns and drug diligence sections.
Biomedical literature citations used in evidence bundles and conversational follow-ups.
Full-text aware literature layer alongside PubMed for broader retrieval.
Open scholarly graph for works, institutions, and concepts when the evidence router includes it.
Public company filings used for commercial / filings–focused Ask modes.
Separate skill portals call ToolUniverse’s tool library (1000+ scientific tools). Those integrations are documented per skill; this table is the core Synthesis corpus only.
Optional MCP / SDK-backed tools for deep research workflows — orthogonal to the Drug Repo ingest list above.
These Cursor skills live under .cursor/skills/. [synthesis.] lists them here; execution happens in Cursor via MCP (uvx tooluniverse) or Python jobs.
Evidence-based guideline retrieval for indication framing and “what is ideal therapy for X?” style questions.
Using clinical guidelines skills, summarize first-line systemic therapy options for moderate-to-severe plaque psoriasis in adults. --- What do major guidelines recommend for induction therapy in biologic-naive Crohn’s disease?
Critique endpoints, eligibility, and statistical strength; pairs with ClinicalTrials.gov context for whether data support claims.
For NCT03066778, comment on whether the primary endpoint and sample size are adequate to support a label-expansion narrative. --- Compare standard-of-care control arms used in recent Phase 3 trials for this indication.
Reverse-style trial discovery and pipeline identification (sponsors, interventions, status windows).
Find actively recruiting trials that include both infliximab-exposed and biologic-naive arms for ulcerative colitis. --- Map competitor trials for the same MOA class as this asset and flag overlapping inclusion criteria.
Disease-level mechanistic and epidemiologic context to complement FDA label rows in the catalog.
Summarize disease burden, standard-of-care biologics, and residual unmet need for hidradenitis suppurativa.
DDI checks especially important for polypharmacy in infusion centers and perioperative pathways.
Assess major interaction risks between infliximab and commonly co-prescribed small-molecule immunosuppressants.
Hypothesis generation for off-label / new-indication expansion (common in autoimmune and rare disease BD).
Given mechanism X, outline repurposing hypotheses supported by real-world usage patterns and genetic evidence.
Broad drug intelligence (targets, class, development history) layered on top of catalog rows.
Produce a structured drug profile for vedolizumab: class, MOA, major indications, and key safety boxed warnings.
Target-level evidence triangulation (genetics, dependency, tractability) for BD / target review sessions.
Summarize genetic and functional evidence supporting TNF as a target in this indication.
Multi-source literature synthesis with citations — default backstop for any complex scientific question.
What does recent literature say about switching between anti-TNF biologics in IBD after secondary loss of response?
Polypharmacology / network views of drug action — useful for combination rationale and class effects.
Outline network-level rationale for combining an anti-integrin with an anti-cytokine in refractory disease.
Safety signal exploration (FAERS-style questions, disproportionality concepts) alongside label text in [synthesis.].
Summarize known infusion reactions and postmarketing safety considerations for this monoclonal antibody class.
Systematic gap analysis + new API/tool proposals for expanding ToolUniverse coverage as this app grows.
Identify gaps in ToolUniverse coverage for infusion-center formulary data sources and propose 3 candidate APIs.
Keeps the skill bundle current when upstream adds new workflows (pairs with DevTU expansion work).
Refresh installed ToolUniverse skills in this repo and summarize what changed upstream.
For scripted pipelines (batch jobs, ETL, CI) calling tools via `ToolUniverse().run(...)` — not required for the Next.js UI.
Sketch a Python script that batches PubMed + OpenTargets pulls for a list of BLA holders from a CSV.
Tracks requested ToolUniverse feature families and current integration posture for Ask/Drug Repo.
Core label retrieval/intelligence path for Ask + Drug Repo enrichment.
Router + agentic orchestration available; used for broad research workflows.
Covered through literature aggregation workflows and citation synthesis.
Planned stronger CDC-specific routing in Ask evidence adapters.
DailyMed class-search usage to be expanded in Ask ranking and filtering.
Already integrated for trial landscape and matching flows.
Must-have deep integration target for Ask and canonical repo curation.
Next stage after ontology-lite scaffold; full mapping not yet enabled.
Available for target → candidate reasoning tracks.
Planned explicit source adapter wiring for DrugCentral-specific coverage.
Regulatory status/generic availability flow already represented.
Planned explicit Ask hook for dynamic tool discovery routing.
Will support aggregate evidence synthesis workflows.
Accessible through ToolUniverse data retrieval stack.
Gene-oriented lookup and interpretation workflows available.
Integrated with canonical ingest and regulatory lookup paths.
Label ingestion + label-derived assertions are active.
Now running in reliable throttled mode with key-backed canary validation.
Ready for rare-disease coverage expansion.
Literature pipelines already include PubMed-class evidence retrieval.
Remote collaboration connector tracked as TBD for later sourcing/codelog follow-up.