HACK_002: Signal From Tomorrow, HOIV Vienna, 26–27 September 2026
Three science tracks, and twenty-four hours to build one of them.
Amass is a partner of HACK_002, the hackathon running under the line “building the future we were promised”, and every team gets $500 in Amass API credits. Amass is one REST API over six cross-linked *Cores: 43M+ papers, 1.2M+ clinical trials, 22K+ molecules, 43K+ genes, FDA and EMA authorizations, and patents. Every answer arrives with the record it came from.
You do not need to be a developer to use any of it, and you do not need a biology degree to enter Block B. Each of the three tracks below opens with a picture of what a finished entry looks like, then three concrete places to start, then a prompt you can paste straight into Cursor.
Doors open at 12:00 on Saturday, building runs from 13:00 to 13:00, and judging follows lunch on Sunday. The venue is HOIV, Arsenalstraße 11, 1030 Vienna, with 24-hour access. Registration and the full schedule are on Luma and the Telos Circle event page. Questions for us: hello@amass.tech.
Pick a track
Which one is yours
Choosing is the first and largest decision of the weekend, so here are the three side by side. Nothing stops you from moving later, but the entries that land tend to be the ones that picked at 13:00 on Saturday and never looked up.
AI for Longevity & Aging
- Choose this if
- You like working with real data, you can read a statistics table, or you already have opinions about ageing biology.
- You will build
- A tool, model, analysis, or hypothesis about ageing and healthspan.
- Judged on
- Held-out error, or an effect estimate with its uncertainty. Explain the sampling and the confounding, and say why the result does not establish a treatment benefit.
Neural Systems
- Choose this if
- You like dynamical systems, graphs, or simulation, and you would rather watch something behave than fit something to a table.
- You will build
- A simulation or an experiment around neural and brain-inspired systems.
- Judged on
- Pattern recall, response delay, network stability, or another metric you define, measured across a parameter sweep and against a baseline.
AI for Scientific Discovery
- Choose this if
- You like building tools and agents, and you are more interested in how research gets done than in any single result.
- You will build
- A system that speeds up one stage of the scientific process.
- Judged on
- Citation support, reviewer agreement, false positives, or time saved, against a manual or simple baseline. A generated hypothesis is not a validated discovery, and saying so in the demo counts in your favour.
A note on judging
Block B is judged by Constantin Convalexius (Medical University of Vienna, Harvard) and Lucas Mair (Technical University of Munich, ProviGen), alongside the general panel. They will have seen the underlying literature before. The Measure line on each card above is the closest thing to a published rubric, so treat it as one.
AI for Longevity & Aging
Biomarkers, ageing clocks, biological data, literature, and the research workflows around them. Public datasets are plentiful here, so this is the track with the shortest distance between sitting down and having a first result on screen.
How it is measured
Held-out error, or an effect estimate with its uncertainty. Explain the sampling and the confounding, and say why the result does not establish a treatment benefit.
Which ageing clock says what
4 studies · 5 clocks · JuFo 2+| Finding | Horvath | Hannum | PhenoAge | GrimAge | DunedinPACE |
|---|---|---|---|---|---|
| IPF lung tissue vs healthy controlsacceleration tracks disease severityAm J Physiol Lung Cell Mol Physiol 2025JuFo 2PMID 39970931 | ↑ | ↑ | ↑ | — | ↑ |
| Fast vs slow glaucoma progressionHorvath +2.93 y, Hannum +1.24 yOphthalmology 2025JuFo 3PMID 39716635 | ↑ | ↑ | ns | ↑ | — |
| Blood vs brain-MRI age, twin cohortheld up in within-pair modellingBrain 2025JuFo 3PMID 39992867 | ns | ↑ | ns | ↑ | ns |
| Drug-naive first-episode schizophreniadeceleration, p = 0.01BMC Psychiatry 2023JuFo 2PMID 36650462 | ↓ | ns | ns | — | — |
↑ accelerated↓ deceleratedns not significant— not reported
Read the Horvath column downwards. It runs up, up, ns, down across four studies, so the clock you pick is a choice you have to defend.
Three places to start
Age-stratified biomarker panel
Pick three biomarkers and plot them by age decade in NHANES. Write down every exclusion you made and say plainly what the sampling cannot tell you. The write-down is the hard part and the part that scores.
NHANES · half a dayOne signature, two studies
Take a single ageing-related expression signature and compare it across two GEO studies. Then find the batch effect and show it, before a judge asks whether you looked.
GEO · one eveningEvidence map of a mechanism
Link a proposed ageing mechanism to the papers behind it and mark the places where they contradict each other. This is the one in the window above, and it is the least code of the three.
BioMedCore · a few hoursPaste this into Cursor to start
Build me an evidence map for one ageing question.
Question: does epigenetic age acceleration track disease severity,
and do the different clocks agree with each other?
Use the Amass API at api.amass.tech with my key in AMASS_API_KEY.
Search BioMedCore for studies since 2023, keep only journals with
journalQualityJufo 2 or higher, and for each paper pull out which
clocks were measured and which direction each one moved.
Lay it out as a matrix with one row per study and one column per
clock, put the PMID on every row, and mark every cell where two
studies disagree. Do not average the disagreements away.The call behind the picture
curl "https://api.amass.tech/api/v1/cores/biomedcore/records\
?query=epigenetic+clock+DNA+methylation+biological+age+acceleration\
&minJournalQualityJufo=2&minPublicationDate=2023-01-01&limit=6" \
-H "Authorization: Bearer amass_YOUR_KEY"TITLE JOURNAL DATE JUFO CITED
Epigenetic age acceleration in idiopathic pulmonary… Am J Physiol Lung 2025-03-01 2 10
Accelerated Epigenetic Aging Is Associated with Fast… Ophthalmology 2025-05-01 3 9
Epigenetic age acceleration in peripheral blood corr… Brain 2025-08-01 3 5
Extracellular vesicles and epigenetic aging clocks i… J Gerontol A 2026-08-04 3 0
Low birthweight is associated with epigenetic age ac… Evol Med Public Hlth 2023-01-01 2 15
Epigenetic clock analysis of blood samples in drug-n… BMC Psychiatry 2023-01-17 2 13Starter data for this track
Also try
GeneCore turns an ageing target into a defensible pick in one call. SIRT1, FOXO3, or whatever your hypothesis rests on comes back with its tractability, its genetic constraint, and its curated safety signals, each with the number behind it. That is the paragraph in your write-up that says why this gene and not another one.
Neural Systems
Neural simulation, in-silico nervous systems, connectomics, and emergent behaviour. This is the track where a small, complete thing beats a large, half-wired thing: a hundred neurons that demonstrably learn one pattern is a finished entry.
How it is measured
Pattern recall, response delay, network stability, or another metric you define, measured across a parameter sweep and against a baseline.
Model card · the assumptions in your network
SCN1A · Nav1.1LOEUF
0.098decile 0pLI
1.00loss intolerantDependent lines
0 / 1,244not essentialPDB entries
7DTD3D solvedWhat it is
Nav1.1, the pore-forming alpha subunit of a voltage-gated sodium channel. 2,009 aa, 229 kDa, four homologous domains of six transmembrane regions each.
Kinetics you can set
Phosphorylation at Ser-1516 by PKC slows inactivation and reduces peak sodium current. That is a knob in your model with a measurement behind it.
What breaking it does
Variants cause Dravet syndrome, generalized epilepsy with febrile seizures plus, and familial hemiplegic migraine 3. Your perturbation has a clinical phenotype attached.
ENSG00000144285UniProt P354982q24.3
Your network, trained
Weights shuffled
The two panels are yours to produce. The card above them is the part you do not have to guess at, and it is the part a judge can check.
Three places to start
A hundred neurons that learn one pattern
Train a small spiking network on a single pattern, then shuffle the synaptic weights and run it again. The comparison against the shuffled control is what turns a demo into a result.
Brian2 · one eveningMotifs under edge removal
Take a small connectivity graph, delete a chosen set of edges, and show how the motif statistics move. Say which edges you chose and why before you show the plot.
Allen Brain Map · half a dayA sensory-to-motor loop that degrades
Wire a simulated loop from sensing to movement, then raise the noise and the signal delay and plot where the response stops being stable. Name the breaking point.
simulation · a few hoursPaste this into Cursor to start
Help me build a small spiking network and a real baseline.
Goal: a 100-neuron spiking network in Brian2 that learns one
input pattern, plus a control with the synaptic weights shuffled.
Set it up so I can sweep one parameter and plot pattern recall
against that sweep, with the control on the same axes.
Give me a spike raster for both conditions.
Then write the assumptions section for me: list every biological
parameter I have hard-coded, and mark which ones I picked out of
convenience rather than from a measurement.The call behind the picture
curl "https://api.amass.tech/api/v1/cores/genecore/records\
?query=SCN1A&include=protein&limit=1" \
-H "Authorization: Bearer amass_YOUR_KEY"symbol SCN1A ENSG00000144285 2q24.3 2,009 aa 229 kDa
targetClass Ion channel › Voltage-gated ion channel › Voltage-gated sodium channel
constraint LOEUF 0.098 pLI 1.00 (LOEUF decile 0) missense o/e 0.564
essentiality not essential, 0 dependent lines of 1,244 screened
structure 3D: yes · PDB 7DTD · Pfam PF00520, PF24609, PF06512, PF11933
ptm Ser-1516 phosphorylation by PKC slows inactivation, lowers peak current
disease Dravet syndrome · GEFSP2 · familial hemiplegic migraine 3Starter data for this track
Also try
Amass does not simulate anything, and it is not meant to on this track. What it does is settle the assumptions argument. One GeneCore call tells you what is actually measured about the channel or receptor your model leans on, so the assumptions slide cites a record instead of a memory.
AI for Scientific Discovery
Knowledge synthesis, scientific agents, experiment design, lab automation, and automated analysis. The trap on this track is building something that sounds right; the entries that land are the ones that show their sources and show where they failed.
How it is measured
Citation support, reviewer agreement, false positives, or time saved, against a manual or simple baseline. A generated hypothesis is not a validated discovery, and saying so in the demo counts in your favour.
Claim check, awaiting a human
4 records · 2 flagsClaim under test
“Lecanemab is approved for early Alzheimer’s disease in the United States and the European Union on the same terms.”
- ✓RegulatoryCoreLeqembi · FDAAMRC_PxJ6…
BLA · active · authorized 2023-01-06 · Eisai Inc · initiate at MCI or mild dementia
- ⚑RegulatoryCoreLeqembi · EMAAMRC_Nvtg…
centralised · active · authorized 2025-04-15 · Eisai GmbH
The EU indication is restricted to ApoE e4 non-carriers and heterozygotes. The US one is not. The claim of identical terms fails here.
- ⚑TrialCoreClarity AD · NCT03887455AMTC_F4vF…
Phase 3 · Eisai · N = 1,906 · active, not recruiting · completion 2029-06-30
hasResults is false. The 18-month outcome everyone quotes is in the literature, not in this registry record. A tool that reads only the registry reports nothing here.
- ✓TrialCoreStudy 201 · NCT01767311AMTC_Jux8…
Phase 2 · N = 856 · completed 2024-12-10 · results posted
The value is not the answer. It is that every line carries the record it came from, and that the tool says where it is not sure.
Three places to start
An evidence graph for one question
Answer a single research question with traceable papers, and make the tool state out loud which claims it cannot support. The refusals are the interesting output, not the answer.
BioMedCore · one eveningAn experiment planner
Propose the controls and a cost-aware design for one real lab question, then put it in front of somebody who runs that experiment and record what they changed.
half a dayA QC notebook that logs its decisions
Find the outliers in a public assay table and write down every flagged row with the reason it was flagged. A reviewer should be able to overturn any one of them.
a few hoursPaste this into Cursor to start
Build me a claim checker with a review step.
Input: a claim about a drug, like "lecanemab is approved for early
Alzheimer's disease in the US and the EU on the same terms".
Use the Amass API at api.amass.tech with my key in AMASS_API_KEY.
Search RegulatoryCore for every FDA and EMA authorization of that
substance, and TrialCore for the trials behind it.
For each record, say whether it supports or contradicts the claim,
and quote the specific wording that decides it.
Flag anything you are not sure about instead of resolving it, and
refuse to output a verdict until a human has cleared the flags.
Show me the failures as prominently as the passes.The call behind the picture
curl "https://api.amass.tech/api/v1/cores/regulatorycore/records\
?query=lecanemab&limit=4" \
-H "Authorization: Bearer amass_YOUR_KEY"NAME AGENCY STATUS AUTHORIZED HOLDER SUBSTANCE
Leqembi FDA ACTIVE 2023-01-06 Eisai Inc Lecanemab-irmb
Leqembi Iqlik FDA ACTIVE 2025-08-29 Eisai Inc Lecanemab-irmb
Leqembi EMA ACTIVE 2025-04-15 Eisai GmbH Lecanemab
Kisunla EMA ACTIVE 2025-09-24 Eli Lilly Nederland B.V. Donanemab
# the EMA indications restrict to ApoE e4 non-carriers and heterozygotes.
# the FDA one does not. that difference is the whole finding.Starter data for this track
Also try
Cross-Core links are the shortcut on this track. A gene resolves to the drugs that target it, those to the trials that ran them, those to the papers behind them and the FDA and EMA labels that cover them. Follow the reference fields one call at a time and every hop in your graph arrives with its own id. Treat an empty reference array as “none recorded”, not as “no evidence”.
The offer
Claim your team’s $500 in API credits
One code per team, worth $500. Sign up with the email you registered for the hackathon with, then ask us for the code.
What $500 buys
Roughly 10,000 searches or 50,000 record fetches. Search costs $0.05 per 20 results; get-by-id and lookup are $0.01 each. That is far more than a team gets through in 24 hours, so query freely and cache what you re-run.
Start here
Four ways in, one key
All four reach the same data with the same key. If you do not know which one you want, start with the assistant and let it write the first call with you.
Start with this
The Amass build assistant
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 completed trial of an amyloid antibody with posted results and the papers behind them. It answers with the endpoint, the filters, and the field names, so you are not guessing at parameter names from a schema you have not read.
The platform
Your account, credits, and API keys, plus the full documentation, the interactive API reference, and the app gallery. Everything the assistant reads from lives here.
platform.amass.techThe Amass app
The product itself. Ask a research question and get an answer with the records attached. Use it on Saturday morning to 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. It also gives you a judge-friendly demo with nothing to deploy.
Connect the MCPThe data
Six *Cores, one API
Same base URL, same auth header, same error shape across all of them. A paper, a trial, a molecule, a gene, an authorization, and a patent can resolve to the same entity, so you 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
Apps to fork
You do not have to start from an empty file
Every app in the gallery runs on the same API, the same key, and the same cited records as the tracks above. All of them were built with Lovable, so there is a route through this weekend 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?”
When to meet
How to reach us over the 24 hours
We are not at the venue this time. We are on the event Discord for the whole weekend, and we run a mentor session on Saturday.
Mentor session
We run a mentor session at 16:00. Bring the question you are stuck on, even if the build is still a sketch.
The rest of the weekend
We are remote, and in Discord for the full 24 hours, or at hello@amass.tech. Questions about credits, endpoints, or which *Core holds what are all fair game.
Playbook
Six things worth knowing before 13:00 on Saturday
These are what separate an entry that lands from one that spends the weekend fighting its own plumbing.
Narrow until it feels too small.
Twenty-four hours is the real constraint, and every track’s brief opens by asking for one narrow, testable question. A finished small thing wins against an unfinished large one, and the judges can tell the difference from the back of the room.
Build the baseline first.
Every Measure line in Block B asks for a comparator. Shuffled weights, a random classifier, the manual version of the workflow. The baseline takes twenty minutes and it is what turns your number into a result.
Show where it fails.
Two of the three track briefs ask for failures explicitly. A demo that names its own failure mode reads as finished work. A demo that hides it reads as a demo that has not been tested.
Cite it or drop it.
Every Amass record carries a stable id and a route back to its source, so 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.
Let the build assistant write the calls.
Describe what you want in plain language and it answers with the endpoint, the filters, and the field names, straight from the documentation. That removes the part of the weekend where you guess at parameter names from a schema nobody has read.
Read the envelope, 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; on a 429, read Retry-After and back off. There is no paging, so narrow with filters instead.
Prototype in Claude
Sketch it in a chat before you write any app code.
Connect the Amass MCP in Claude, Cursor, ChatGPT, or Codex and query all six *Cores in natural language. Explore the data, find the filters that work, and only then drop the same queries into your build. On Sunday it doubles as the demo: the answers come back cited in the chat, with nothing to deploy.
In Claude Code, install the Amass skill and let the model wire up your data layer for you. That is a short path from “what is even in here?” to a working prototype, which is the right question to answer on Saturday afternoon rather than Sunday morning.
Pick one, and go build it
Claim your team’s $500, choose a track, and let the API do the reading. Narrow it until it feels too small, build the baseline, and show where it breaks. We would like to see it on Sunday.
One code per team. Ask in the event Discord or at 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