Luvia — Walk-through

01 Thesis

The hard part
isn't the AI.

An LLM can summarize, reason, and write — it can search the web. What it still can't do, an analyst can. And that's the problem we solve.

Know what's true

Web search returns press releases and outdated articles. A curated, cross-referenced database of 827,000 drug programs returns signal.

Cross-reference

MK-3475, pembrolizumab, and Keytruda are the same molecule. Identity resolution across 56 identifier systems makes it obvious.

Query at scale

An LLM looks up one drug at a time. We run structured queries across 41M molecules, 264M patent links, and 577k trials.

Show its work

Every answer traces back to a primary source — a specific database record, a specific trial ID, a specific patent family. No hand-waving.

The AI is the interface.
The value is in the infrastructure underneath.

01 — thesis

02 What we maintain

A data layer,
not a search engine.

We maintain the substrate — curated, cross-referenced, refreshed. Ask a question in plain English; the system decides what to query.

Drug programs 827K Every publicly-disclosed program, with company, target, indication, phase, and regulatory status.

Molecules 41M Structures, properties, identifiers, parent/salt resolution — reconciled into one record.

Patent links 264M Molecule-to-patent links with evidence tiers, expiry, jurisdiction, family.

Clinical trials 577K The full ClinicalTrials.gov corpus, linked to molecules, targets, and sponsors.

On top of the substrate — 25+ live API connections.

Routed on demand

Literature

PubMed · PMC · OpenAlex · Semantic Scholar · bioRxiv

Regulatory & safety

OpenFDA · EPO · USPTO

Targets & biology

OpenTargets · UniProt · STRING · InterPro · IUPHAR

Structural

AlphaFold · PDB · PubChem · ChEMBL

Expression & genomics

GTEx · GWAS Catalog · Allen Brain Atlas · Antibodypedia

Pathways & PK

Reactome · Pathway Commons · PharmGKB · PKDB

03 How it feels to use

Ask once.
Get an answer.

Target, indication, phase, status, sponsor — assembled into a clean competitive table. Not a list of links.

vault / kras-g12c-nsclc-landscape.md

ASK

What's the competitive landscape for KRAS G12C inhibitors in NSCLC?

Asset

Sponsor

Phase

Line

ORR

Key trial

SotorasibAMG 510

Amgen

Acc. Appr.

37%

CodeBreaK 100

AdagrasibMRTX849

BMS (Mirati)

Acc. Appr.

43%

KRYSTAL-1

OlomorasibLY3537982

Eli Lilly

Ph3

63%

SUNRAY-01

DivarasibGDC-6036

Roche

Ph3

56%

Krascendo-1

DaraxonrasibRMC-6236

Revolution Med.

Ph3

38%

RASolve 301

BI 1823911—

Boehringer

Ph1

—

NCT04973163

6 programs · 6 sponsors · sourced from OpenTargets, ClinicalTrials.gov 0.41s

04 The conversation continues

Follow-ups that compound.

Each step builds on the last — context carries forward, data re-joins, output sharpens.

Add IP exposure

“Which of these have patent families expiring before 2029?”

Joins 264M molecule–patent links. Returns expiry date, jurisdiction, and family.

Enrich with live reads

“Pull the latest readouts and publications for the Phase 3 programs.”

Live hits from ClinicalTrials.gov and PubMed, filtered to the sponsors above.

Analyse & rank

“Summarise as a competitive memo — rank by clinical risk, flag IP cliffs, identify the most differentiated program.”

Not a summary. A judgment — with sources, ready to paste into an IC memo.

The conversation doesn't reset.
Neither does the context.

04 — continuity

05 Your workspace

Research that accumulates.

Every memo, table, and scan is saved — and linked, automatically, to the entities it touches. Future questions arrive pre-contextualised. Month three, the system anticipates them.

~/vault

kras-g12c-nsclc-landscape.mdApr 18
patent-expiry-analysis.mdApr 18
phase3-literature-review.mdApr 19
competitive-memo-2026-04.mdApr 19
gsk-oncology-pipeline.mdApr 12
adc-payload-linker-scan.mdApr 08
obesity-glp1-landscape.mdMar 31
weekly-briefing-2026-w16.mdToday

06 Positioning

A different kind of tool.

Static intelligence tools look up known entities. Dynamic research handles the analytical work. Both have a place.

	Static intelligence	Dynamic research / Luvia
Interface	Search, browse, filter	Ask in natural language
Output	Pre-built profiles & reports	Tables, charts, memos shaped to the query
Cross-referencing	Manual — stitch in Excel	Automatic — 56 identifier systems
Freshness	Editorial update cycle	Live APIs, on demand
Follow-up	New search, start over	Conversation continues, builds on prior
Accumulation	Bookmarks and exports	Persistent workspace that compounds
Customisation	Choose your modules	Shaped to your team

07 The boundary

We do one thing.
Make the analytical work faster and sharper.

Where Luvia fits

Due diligence. Competitive landscapes. Patent and IP exposure. Target-indication analysis. Portfolio monitoring. Live readouts filtered to what you actually work on.

The hard partisn't the AI.