BETABiohacksAI is a research tool for informational purposes only. All outputs are computational hypothesis candidates — not confirmed mechanisms, not medical advice, and not a substitute for professional medical judgment. Independent experimental validation is always required.
BiohacksAI is an evolving scientific literature platform. New compounds and evidence are indexed continuously.
B
B
BiohacksAI
Evidence-Based Biohacking Platform

How BiohacksAI works

BiohacksAI is an AI-driven compound research and discovery platform built on 1.14 million PubMed studies, a deterministic graph of 31,590 substances, and a cryptographic evidence chain called X-Vault.

The platform does two things: it lets researchers query the scientific literature with evidence-grade answers, and it systematically discovers novel compound–pathway combinations that may have research value.

31,590
Substances
3,120
Targets
2,209
Pathways
1.14M+
PubMed studies
01 — Corpus

The evidence layer

Every claim in BiohacksAI is backed by a specific PubMed study. The corpus is a curated, versioned snapshot of PubMed abstracts indexed by substance, target, pathway, and study type.

The corpus is locked at a specific version — currently v20260312-01. When you query the system you always get an answer from the same dataset, and every answer includes a Merkle root hash that proves which version was used.

Determinism — why it matters
The same question asked today and in six months will produce the same answer from the same corpus version. The result is reproducible and cryptographically bound to the data it came from. This is not how standard AI chatbots work.
Root hash
85b6464f…c8efb
Corpus version
v20260312-01
Reproducible
Yes — always
02 — Discovery Engine

How discoveries are generated

The discovery engine scans the corpus for compounds that have strong biological target coverage but low literature footprint — meaning the scientific community has not yet fully characterised them.

These candidates are ranked by a two-layer scoring system and sealed cryptographically into the X-Vault evidence chain.

DISCOVERY_v1.1 — six-step pipeline
1
Filter candidates
31,590 graph entities filtered: discovery_score > 40, studies < 20, targets ≥ 3, pathway diversity ≥ 2.
CACHE.underratedIndex → filterCandidates()
2
Deduplicate against vault
Each candidate gets a discovery_hash. If that exact compound × target × pathway × corpus_version combination is already sealed, it is skipped — discoveries are never duplicated.
sha256(compound + sorted_targets + sorted_pathways + corpus_version)
3
Enrich
Top molecular targets and biological pathways are resolved from the graph. Biotech score is computed. High-value target hits are identified from the HIGH_VALUE_TARGETS registry.
computeBiotechScore() · HIGH_VALUE_TARGETS
4
Seal into X-Vault
The discovery event is appended to the WORM chain. An event_hash and chain_hash are generated. This step is permanent and immutable.
discovery-seal.js → WORM JSONL
5
Generate PDF
A cryptographically sealed PDF report is generated: mechanism hypothesis, targets, pathways, X-Vault seal, validation roadmap.
6
Write ledger
A JSON ledger entry is written. This is the source of truth for the marketplace listing.
data/discoveries/disc_YYYYMMDD_NNN.json
Nightly automation
The pipeline runs automatically every night at 02:00 UTC, sealing up to 50 new discoveries per run. A compound will not appear twice for the same pathway context against the same corpus version.
03 — Scoring & Plans

Discovery score and access tiers

Every compound is evaluated by two complementary scores. The discovery score measures biological breadth. The biotech score measures research relevance against known drug targets.

Discovery Score — biological breadth
target_count × pathway_diversity / log(studies + 2)

Compounds with many validated targets, broad pathway coverage, and low literature footprint score highest. The log-damping on studies ensures that well-studied compounds are de-prioritised in favour of understudied ones.

Biotech Score — research relevance (DISCOVERY_v1.1)
(target_relevance × pathway_diversity) / log(studies + 2)

Target relevance is a weighted sum based on the HIGH_VALUE_TARGETS registry — 60+ gene symbols classified as Tier 3 (kinases, transcription factors, metabolic regulators: EGFR, NFE2L2, MTOR, SIRT1, TP53, BRD4, JAK1…) or Tier 2 (research-grade: MAPK1, SOD1, APP, HDAC2…).

Subscription plansCOMING SOON
Explorer
~$19/mo
  • ✓ All compound profiles
  • ✓ Target + pathway pages
  • ✓ BiohackerPanel
  • ✓ Stack Lab (hypothesis)
  • ✓ Ayurveda search
  • — Discoveries marketplace
  • — API access
Researcher
~$49/mo
  • ✓ Everything in Explorer
  • ✓ Discoveries marketplace
  • ✓ White Space analysis
  • ✓ Knowledge graph
  • ✓ Export (CSV / JSON)
  • ✓ API — 1,000 req/day
  • — Full API
API / Enterprise
Custom
  • ✓ Everything in Researcher
  • ✓ Full API — unlimited
  • ✓ Discovery report access
  • ✓ Bulk compound queries
  • ✓ Webhook on new discoveries
  • ✓ Corpus version pinning
  • ✓ Priority support

Pricing and availability will be announced. Join the waitlist: info@eveverified.com

04 — X-Vault

Cryptographic evidence sealing

Every discovery is sealed into the X-Vault — an append-only, SHA-256 hash-chained log that provides a cryptographically verifiable record of when a discovery was made and what corpus version it was based on.

Event hash
SHA-256 of the complete discovery event in canonical JSON. Proves the content has not been altered.
Chain hash
SHA-256 of (previous chain hash + event hash). Links every discovery to all prior discoveries — tampering breaks the chain.
Discovery hash
Deduplication key: sha256(compound + sorted targets + sorted pathways + corpus version). Prevents re-sealing identical discoveries.
Vault index
Sequential integer. #1 came before #2. Order is immutable and independently verifiable.

The X-Vault WORM log is stored server-side and is never rewritten — only appended to. Anyone can independently verify a discovery at biohacksai.com/verify.

05 — Mechanism Graph

Visualising how a compound works

Every discovery includes a mechanism discovery graph — a force-directed visualisation showing the relationship between the compound, its molecular targets, and the biological pathways those targets regulate.

Three-ring graph layout
Centre · Compound

Fixed at centre. The compound being analysed.

Ring 1 · Molecular targets

The gene/protein targets the compound interacts with. Colour-coded by research relevance: gold = Tier 3 (kinases and TFs), indigo = Tier 2 (research targets), teal = standard. A pulsing dashed ring marks Tier 3 targets with ≤ 5 studies — the knowledge gap indicator.

Ring 2 · Biological pathways

The biological systems those targets belong to (Reactome). Shown as dashed rectangles connected by dotted lines from their associated targets.

06 — Marketplace

Discovery reports

Sealed discoveries are listed in the marketplace. Each entry shows the compound, mechanism pathway, discovery score, and X-Vault seal. Contact us to request access to the full report.

07 — Philosophy

AI proposes. Humans decide.

BiohacksAI is built on the EVE (Evidence & Verification Engine) architecture developed by Organiq Sweden AB. The core principle is witness mode: the AI generates hypotheses and surfaces evidence, but all decisions are made by humans and recorded with human-controlled cryptographic keys.

What witness mode means in practice
  • — The discovery engine never licenses anything autonomously
  • — Every discovery seal requires a verified admin key
  • — The system logs what it proposed and when — the human logs what they decided
  • — No discovery can be unsealed or retroactively altered

All discoveries are labelled as computational hypothesis candidates — not confirmed mechanisms. Independent experimental validation is always required.

08 — Verification

Verify any discovery

Anyone can verify a discovery — no account required. Paste a Discovery ID or event hash into the verification page. The system walks the X-Vault chain and confirms whether the discovery exists and whether the chain integrity is intact.

10 — Self-Growing Corpus

The platform that grows while you sleep

Most research databases are static — they reflect the world as it was when someone last updated them. BiohacksAI is different. Every Sunday at 03:00 UTC, the compound discovery engine runs autonomously: scanning PubMed, extracting candidates, validating against PubChem, and promoting new substances into the evidence corpus — all without human intervention.

Weekly autonomous cycle — what happens at 03:00 UTC every Sunday
1
Scan 23 biological targets
The engine queries PubMed across 23 longevity, cognitive, and cellular targets — AMPK, SIRT1, mTOR, NRF2, BDNF, ULK1, TFEB, GSK3β, HIF1A, PPARα and more — fetching up to 200 articles per target.
23 targets × 200 articles = ~4,600 articles per run
2
Three-layer chemical filter
Every MeSH term extracted from those articles passes three independent filters: BIO_STOP (120+ biological non-compound terms rejected), ChemPattern (regex for chemical naming conventions), and PubChem HTTP validation. Only real, validated chemical compounds survive.
L1: BIO_STOP → L2: ChemPattern → L3: PubChem
3
Threshold and promote
Candidates appearing in 5 or more articles are automatically promoted to the evidence corpus. Each promotion is logged with a SHA-256 cryptographic seal, timestamped, and written to the immutable corpus-promotions ledger.
AUTO_PROMOTE_THRESHOLD = 5 · xvaultSeal(SHA-256)
4
Re-index and ingest
For every newly promoted compound, the engine fetches up to 200 PubMed articles dating back to 2010 — building a full evidence profile. The knowledge graph is updated with new target and pathway relationships.
cycle-ingest.js → +1,000 articles per run (avg)
5
New Merkle root
Once ingestion is complete, a new Merkle root is computed over the entire corpus. This cryptographic fingerprint proves exactly what evidence existed at that moment — reproducible, auditable, tamper-evident.
SHA-256 Merkle tree → new corpus_version
What this means in practice
When the corpus was first built, it contained 139 substances. After one autonomous weekly run, it grew to 147 — adding Silymarin, Ergothioneine, Ergosterol, Stigmasterol, Vitamin B12, Riboflavin, and alpha-Linolenic Acid, each with a full PubMed evidence profile and cryptographic promotion record. No human decided to add them. The evidence did.
Targets scanned
23
Articles per run
~4,600
Promotion log
X-Vault sealed
11 — Data Attribution

Sources & licenses

BiohacksAI is built on publicly available scientific databases. We are grateful to the teams behind these resources.

SourceUsageLicense
PubMed / NCBILiterature corpus (1.14M abstracts)Public domain (NLM)
ChEMBLBioactivity data, targets, assaysCC BY-SA 3.0
BindingDBBinding affinity dataCC BY 3.0
COCONUTNatural product compounds (~370k substances)CC BY 4.0
ReactomePathway graph (2,209 pathways)CC BY 4.0
UniProtProtein / target metadataCC BY 4.0
PubChemCompound identifiers (CIDs)Public domain (NIH)

Data is used for non-commercial research and hypothesis generation. Bioactivity records from ChEMBL and BindingDB are aggregated and scored; individual assay records are not redistributed. Natural product data from COCONUT is used for compound identification only. If you believe any usage conflicts with a source license, contact info@eveverified.com.

🚧 Under Construction

BiohacksAI is currently in active development and open to early access. The platform will be locked to approved users when we launch — sign up now to secure a spot.

Demo licence — Email info@eveverified.com to request early access.

Become a development partner — Interested in co-developing BiohacksAI or joining as an early partner? We are open to conversations with researchers, institutions, and investors. Reach out at info@eveverified.com.