AI in Life Sciences, DTU Skylab, Wed 16 September 2026
Turn life-science data into software in one evening.
Amass hosts this hackathon with Cursor Copenhagen and DTU Skylab. One REST API covers the six cross-linked *Cores on the public API: 43M+ papers, 1.2M+ trials, 22K+ molecules, 43K+ genes, FDA and EMA authorizations, and patents. Every answer traces back to the record it came from.
Copy the credit code, then pick one of the four tracks. You can also fork one of the apps in the gallery and change the question it asks. The build assistant writes the API calls with you.
Doors open at 14:30 and the kickoff is at 15:00. Building runs from 15:30, demos start at 17:30, and prizes are announced at 18:00. The venue is the DTU Skylab Auditorium at Diplomvej 373D. Email hello@amass.tech if you get stuck, or find us in the room.
Start here
Four ways in
All four give you the same data with the same key. If you are not sure which one you want, start with the assistant.
Start with this
The Amass build assistant
This is a chat grounded in the Amass documentation, built to help you write against the API. Describe what you are trying to do, such as finding every Danish Phase 2 trial in atopic dermatitis and the papers behind them. It answers with the endpoint, the filters and the field names, so you do not have to guess parameter names from a schema you have not read.
The platform
This is where your account, credits, and API keys live, alongside the full documentation, the interactive API reference, and a gallery of apps you can fork. Everything the assistant reads from is here.
platform.amass.techThe Amass app
This is the product itself. Ask a research question and get an answer with the records attached. Use it to scope an idea and see what the data can answer before you write a line of code.
preview.amass.techMCP
Connect Amass to Claude, Cursor, or Codex and query all six *Cores in natural language while you build. Use it to find the filters that work before you hard-code them.
Connect the MCPApps to fork
You do not have to start from an empty file
Every app in the gallery runs on the same REST API, the same key, and the same cited records as the tracks below. All of them were built with Lovable, so there is a route through this that involves no local setup at all. Fork one, point it at your key, and change the question it asks.
From Gene to Approval
One gene → every drug, trial, and FDA/EMA approval behind it.
Graveyard & Garden
25 years of late-stage trials, halted or still alive.
Clinical Trial Atlas
Where a drug is tested, and which regions are enrolling.
Indication Timeline
A drug across its Phase-3 indications and approvals.
Transatlantic Divergence Desk
FDA versus EMA outcomes, side by side.
Pipeline Monitor
A weekly digest of new Phase 2/3 trials by sponsor.
Nineteen starter apps across research, regulatory, and clinical. Each one is a working answer to “what does a small app on this data look like?”
The data
Six *Cores on the public API
These are the six *Cores that the REST API and the MCP serve today. They all take the same base URL and the same auth header, and they all return errors the same way. A paper, a trial, a molecule, a gene, an authorization and a patent can resolve to the same entity, so you can follow one thread instead of joining three exports by hand.
BioMedCore
43M+PubMed, PMC and conference abstracts
/cores/biomedcore/records
TrialCore
1.2M+Clinical trials indexed
/cores/trialcore/records
DrugCore
22K+ChEMBL drugs & molecules indexed
/cores/drugcore/records
RegulatoryCore
7K+FDA & EMA documents indexed
/cores/regulatorycore/records
GeneCore
43K+Human genes indexed
/cores/genecore/records
PatentCore
170M+Life-science patents indexed
/cores/patentcore/records
The tracks
Four tracks to choose from
Pick one and make it yours. The first three come with a prompt you can paste into Cursor to get started, so you do not have to write the first line yourself. Every response shown below is a real record from the live API, trimmed to fit the column.
Review the literature in your niche
Pick a topic you care about and let an agent read everything published since you last looked. It keeps only the journals you would actually cite, and it writes a short digest with a PMID on every line. The version worth building is the one that runs again next Monday without you.
Monday digest · engineered probiotics
5 new since 1 May- Harnessing the microbiome for cancer therapyNat Rev Microbiol2026-06-01JuFo 3PMID 41486395
- Living drug carriers: microbial and bioengineered platforms redefining precision deliveryJ Control Release2026-08-24JuFo 3PMID 42636890
- The gut microbiome and allograft outcomesJ Clin Invest2026-08-17JuFo 3PMID 42606123
- Targeted and quantitatively modeled biologic delivery via dual-functional probiotic yeastJ Control Release2026-06-10JuFo 3PMID 41942055
- Developing spore-forming gut bacteria as model organismsTrends Microbiol2026-08-01JuFo 2PMID 42069481
One filter kept the journals worth citing. Every line carries the PMID it came from, so you can check any of them.
Paste this into Cursor to start
Build me a weekly literature digest.
Topic: engineered probiotics for gut disease.
Use the Amass API at api.amass.tech with my key in AMASS_API_KEY.
Search BioMedCore for anything published in the last 30 days, keep
only journals with journalQualityJufo 2 or higher, and write the
result as markdown with the PMID on every line.
Then save it as a script I can run again next Monday.The call behind it
curl "https://api.amass.tech/api/v1/cores/biomedcore/records\
?query=engineered+probiotic+living+therapeutic+gut+microbiome\
&minPublicationDate=2026-05-01&minJournalQualityJufo=2&limit=5" \
-H "Authorization: Bearer amass_YOUR_KEY"TITLE JOURNAL DATE JUFO
Harnessing the microbiome for cancer therapy Nat Rev Microbiol 2026-06-01 3
Living drug carriers: microbial and bioengineered platf… J Control Release 2026-08-24 3
The gut microbiome and allograft outcomes J Clin Invest 2026-08-17 3
Targeted and quantitatively modeled biologic delivery… J Control Release 2026-06-10 3
Developing spore-forming gut bacteria as model organisms Trends Microbiol 2026-08-01 2Also try
You can point it at yourself instead of a topic. authorOrcids returns your own papers, and minCreateDate tells you what entered the index since your last run. That is the difference between a digest and a diff.
Check prior art before the idea becomes a company
Skylab is full of projects that could be spun out, and the first honest question is what somebody has already claimed. Full-text search runs over the title, the abstract, the claims and the description. It also collapses each patent family to one publication, so a landscape comes back as five inventions rather than twenty filings of the same one.
Precision fermentation · recombinant milk protein
5 families · 4 assignees · US| Perfect Day Inc | 2019-04-22 · 2020-02-19 9 cited-by | |
| Alpine Roads Inc | 2020-09-30 16 cited-by | |
| Re Milk Ltd | 2021-05-12 1 cited-by | |
| Fonterra Cooperative Group | 2021-06-24 0 cited-by | |
| 201920202021 |
This is one row per family, not one row per filing. Twenty filings of the same invention collapse to the five things anyone claimed.
Paste this into Cursor to start
Help me check prior art before we take this idea further.
Idea: making milk proteins by precision fermentation in yeast.
Use the Amass API at api.amass.tech with my key in AMASS_API_KEY.
Search PatentCore, keep US filings, and give me one row per patent
family rather than one per filing. For each row show the assignee,
the priority date and the citation count, sorted oldest first.
Then tell me plainly which of them I have to read.The call behind it
curl "https://api.amass.tech/api/v1/cores/patentcore/records\
?query=precision+fermentation+recombinant+milk+protein\
&countryCode=US&limit=5" \
-H "Authorization: Bearer amass_YOUR_KEY"PUBLICATION ASSIGNEE PRIORITY CITED-BY
US-10988521-B1 ALPINE ROADS INC 2020-09-30 16
US-2023106635-A1 PERFECT DAY INC 2020-02-19 9
US-2024215599-A1 RE MILK LTD 2021-05-12 1
US-2024294589-A1 FONTERRA COOPERATIVE GROUP 2021-06-24 0
US-12458039-B2 PERFECT DAY INC 2019-04-22 0
# one row per family. priorityDate is the prior-art cutoff.Also try
Add include=claims to a hit and you see what is actually claimed rather than what the abstract implies. referencesBiomedCore walks back to the papers behind the filing.
Use AI to analyse a dataset
Bring a dataset and let a coding agent do the first pass over it. Point Cursor at the file, ask what is in it, and make it show the code and the chart behind every claim it makes. Your own measurements work best here, and so does anything you can download in an hour. Remember that a chart is not a finding. Say what the data supports, and say where it is too thin.
Your dataset, worked through in Cursor
4 stepsRead the file and profiled it
Columns, types, row count, missing values, and anything odd.
Plotted what actually varies
One chart per column worth a second look, and nothing else.
Proposed three questions the data can answer
Then said which of the three the data supports best.
Answered the strongest one
With the code it ran and the chart behind the claim.
The data is yours and the agent does the grunt work. Checking the result against published evidence is the optional last step.
Paste this into Cursor to start
Here is my dataset: measurements.csv.
First tell me what is in it. Columns, types, row count, missing
values, and anything that looks wrong.
Then suggest three questions this data can actually answer, say
which one is strongest, and answer that one.
Show me the code you ran and the chart behind every claim, and tell
me where the data is too thin to support a conclusion.Also try
When a result looks interesting, the next question is whether somebody has already found it. That is where the Amass API earns a place on this track rather than at the center of it. One BioMedCore search tells you whether your result is new or well known, and it hands you the PMIDs so a judge can check.
Open track: bring your own problem
Bring the problem you already care about. This track has no shape we can describe in advance, and that is the point of it. One thing does hold across the entries that land, though: they chain several sources into a single answer, and they can show where each part of that answer came from. The example on the right is one shape that takes. Yours does not have to look anything like it.
One question, answered across four *Cores
4 calls · GLP1RLOEUF
0.494decile 1pLI
0.986constrainedPDB entries
703D solvedDependent lines
2 / 1,233not essential- GeneCoreGLP1RENSG00000112164
Family B GPCR · 6p21.2 · 463 aa · small molecule at Advanced Clinical
- DrugCoreSemaglutideCHEMBL2108724
protein · 44 approved indications · Ozempic, Rybelsus, Wegovy
- TrialCoreSELECTNCT03574597
Phase 3 · Novo Nordisk · N=17,604 · completed 2023, results posted
- RegulatoryCoreWegovyAMRC_3XhV…
FDA 2021-06-04 and EMA 2022-01-06 · both authorizations active
Gene, then drug, then trial, then approval. The links come with the records, so no hop in the chain is a guess.
The call behind it
curl "https://api.amass.tech/api/v1/cores/genecore/records\
?query=GLP1R&include=protein&limit=1" \
-H "Authorization: Bearer amass_YOUR_KEY"symbol GLP1R ENSG00000112164 6p21.2 463 aa
targetClass Membrane receptor › Family B GPCR › Glucagon-like peptide receptor
tractability small molecule: Advanced Clinical · other clinical: Approved Drug
constraint LOEUF 0.494 pLI 0.986 (LOEUF decile 1)
essentiality not essential, 2 dependent lines of 1,233 screened
structure 3D: yes · 70 PDB entries · Pfam PF00002, PF02793Also try
include=referencesDrugCore returns the DrugCore ids for the molecules that target a gene. Each of those carries its own trial, literature and authorization links, so one call at a time gets you a long way. Follow the reference fields, and cite every hop.
Playbook
Five things worth knowing before 15:30
These are the things that separate a demo that lands from one that spends the evening fighting its own plumbing.
Two hours is the real constraint.
Pick one endpoint and one output that somebody can see on a projector. A digest that actually runs beats a platform that almost does.
Let the build assistant write the calls.
Describe what you want in plain language. The assistant answers from the documentation and hands back the request, the filters and the field names. That is most of the fiddly part done for you.
Ground every claim in a record.
Each result carries a stable Amass id and a route back to its source. There are no invented PMIDs and no invented NCT numbers. That is the whole reason to query the data rather than ask a model from memory.
Narrow with filters, because there is no paging.
A search returns at most 300 rows. Reach for minJournalQualityJufo, phase, maxConstraintLoeuf, assignee and date windows. A tighter query gives you better results and costs fewer credits.
Read the envelope and respect the limit.
Results arrive under data and failures arrive under error, so check for an error before you parse. The limit is 60 requests per 60 seconds. If you get a 429, read Retry-After and wait before you try again.
Go build something tonight
Redeem the code, pick a track, and let the API do the reading. We would like to see what you made at 17:30.
$50 in API credits
Redeem it at platform.amass.tech once you have created your account. If it does not apply, find us in the room or email hello@amass.tech.
- Data
- No training on your data
- Hosting
- EU-hosted, enterprise
- Isolation
- Tenant-isolated by architecture
- Certification
- ISO 27001, SOC 2, NIS2: in motion