🏗️ MindNetQ Cortex Architecture
Governed AI Drug Discovery Platform · Updated June 20, 2026
CortexQ is an autonomous scientific reasoning agent. It discovers cancer targets, generates novel molecules, validates through GPU docking, and learns from outcomes. MindNetQ governs every decision through authority contraction, evidence chains, and hash-linked audit trails. No action without governance approval.
System Diagram (Current)
┌──────────────────────────────────────────────────────────────────────────────┐
│ MindNetQ (Governance Layer) │
│ Authority Contraction · Evidence Hash Chain · Sentinel · Drift Detection │
└────────────────────────────────────┬─────────────────────────────────────────┘
│
┌────────────────────────────────────▼─────────────────────────────────────────┐
│ CortexQ Reasoning Engine │
├──────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌───────────────┐ ┌──────────────┐ │
│ │ 15 Brains │───→│ 38 Capsules │───→│ CortexQ Gen │ │
│ │ (brain-bus) │ │ (compressed) │ │ v4 (SELFIES) │ │
│ └──────┬──────┘ └───────────────┘ └──────┬───────┘ │
│ │ │ │
│ │ ┌──────────────────┐ │ │
│ └────────→│ Cortex-GEN V3 │←─────────┘ │
│ │ (heuristic) │ │
│ └────────┬─────────┘ │
│ │ │
│ ┌────────▼─────────┐ │
│ │ GPU Docking │ ← Uni-Dock (g4dn.xlarge Spot) │
│ │ cortex-docking- │ │
│ │ gpu:13 │ │
│ └────────┬─────────┘ │
│ │ │
│ ┌────────▼─────────┐ │
│ │ NPEAD v5.2 │ ← 18-dim scoring (advancement+learn) │
│ │ + Gate Learning │ │
│ └────────┬─────────┘ │
│ │ │
│ ┌────────▼─────────┐ │
│ │ Outcomes → │ │
│ │ Brain-Bus → │──→ Rejection Sampling → Retrain │
│ │ Weight Update │ │
│ └─────────────────┘ │
│ │
└──────────────────────────────────────────────────────────────────────────────┘
Discovery Pipeline (end-to-end)
Omics Data (TCGA + DepMap + ChEMBL + STRING + 24-axis enrichment)
│
▼
┌─ physician-rs ─────────────────────────────────────────────────┐
│ Novel target discovery (omics → driver classification) │
│ 10 pancreatic targets discovered autonomously │
└────────────────────────────┬───────────────────────────────────┘
▼
┌─ FRSI + Enrichment ────────────────────────────────────────────┐
│ Regulatory inversion scoring · 24-axis gene enrichment │
│ FRSI is advisory (not a hard gate) · Target justification │
└────────────────────────────┬───────────────────────────────────┘
▼
┌─ Generation ───────────────────────────────────────────────────┐
│ Path 1: Known inhibitors (ChEMBL scaffolds → medchem) │
│ Path 2: FGAO fragments (recombination + expansion) │
│ Path 3: CortexQ Gen v4 (SELFIES, novel targets, 100% valid) │
│ All paths read: capsule intelligence + NPEAD weights + avoid │
└────────────────────────────┬───────────────────────────────────┘
▼
┌─ GPU Docking (Uni-Dock) ───────────────────────────────────────┐
│ Spot g4dn.xlarge · 4 regions · Max 3 jobs/gene │
│ Job def: cortex-docking-gpu:13 · Watchdog monitors 15min │
└────────────────────────────┬───────────────────────────────────┘
▼
┌─ Scoring + Validation ─────────────────────────────────────────┐
│ NPEAD v5.2 (18 dims) · 30 med-chem danger patterns │
│ Chemistry gates · PAINS · Reactivity · Lipinski │
│ Grade: A+ (Vina ≤ -8.5) · A (≤ -7.5) · B+ (≤ -6.5) · B │
└────────────────────────────┬───────────────────────────────────┘
▼
┌─ Evidence Chain + Learning ────────────────────────────────────┐
│ Hash-linked audit trail per hero (562 events, 100% coverage) │
│ Outcomes → brain-bus → NPEAD weight update → capsule refresh │
│ Rejection sampling: 1000 → 50 → 10 → fine-tune model nightly │
└────────────────────────────────────────────────────────────────┘
Infrastructure
| Component | Technology | Deployment | Revision |
|---|
| Python Nightly | ECS Fargate (us-east-1) | EventBridge cron 0 2 UTC | cortex-python-nightly:19 |
| Rust Nightly | ECS Fargate (us-east-1) | EventBridge cron 30 2 UTC | cortex-rs-nightly:75 |
| GPU Docking | AWS Batch Spot (4 regions) | On-demand submission | cortex-docking-gpu:13 |
| Physician-rs | ECS Fargate | Triggered by nightly | sha256:409b918d |
| Next-Action-rs | ECS Fargate | Triggered by nightly | sha256:1e0e91bc |
| API | Lambda + API Gateway | cortex-api | Latest |
| Watchdog | Lambda (15min) | cortex-job-watchdog | Latest |
| Web | S3 + CloudFront | app.cortex.mindnetq.com | 175+ pages |
Data Architecture
| Table | Records | Purpose | Access |
|---|
| cortex-heroes | 110+ | All targets under evaluation | Every nightly + API |
| cortex-brain-bus | 2500+ | Signals, capsules, playbooks, learning | Every nightly + GPU jobs |
| cortex-molecules | 500+ | Docked compounds per hero | GPU writes, API reads |
| edice-platform-state | 200+ | Config, GPU map, weights, pools | All agents |
| fgao-fragment-memory | 233 | Validated drug fragments | GPU generation |
Intelligence Stack
| Layer | Component | What it does |
|---|
| Discovery | physician-rs | Finds novel cancer targets from omics |
| Reasoning | 15 Brains | Genomics, pathway, clinical, chemistry, safety, curiosity, etc. |
| Compression | 38 Knowledge Capsules | Decision-ready claims per brain per cancer type |
| Generation | CortexQ Gen v4 + Cortex-GEN V3 | Novel molecules (SELFIES) + scaffold optimization |
| Validation | Uni-Dock + NPEAD v5.2 | GPU docking + 18-dim authority scoring |
| Learning | Rejection Sampling + Phase 9 | Model self-improves from outcomes every night |
| Governance | MindNetQ + Evidence Chain | Hash-linked audit, authority contraction, sentinel |
| Memory | CIFOG GCN (276 nodes) | Graph neural network predicts target authority |
Evidence Chain (Auditable)
Every decision is hash-linked and tamper-evident.
Coverage: 110/110 heroes (100%) · 562 total events · 8 event types
Hash integrity: ✅ 100% verified (recomputed June 20, 2026)
Playback: Full timeline reconstructable per hero — every chemistry validation, brain sync, CDx refresh, evidence assessment, and gap analysis is recorded with timestamp + decision + detail.
GPU Fleet
| Region | Quota (vCPU) | Instance | Cost Model |
|---|
| us-east-1 | 40 | g4dn.xlarge | Spot only |
| us-east-2 (primary) | 8 | g4dn.xlarge | Spot only |
| us-west-2 | 8 | g4dn.xlarge | Spot only |
| ca-central-1 | 8 | g4dn.xlarge | Spot only |
No on-demand spending without PI approval. Watchdog enforces per-gene dedup (max 3). Single GPU map from DynamoDB.
Nightly Schedule (Python + Rust)
| Time (UTC) | System | Steps |
|---|
| 02:00 | Python (rev 19) | Enrichment → GCN → FRSI → FGAO → EGAO → Repurpose → CDx → Chemistry Validation → Physics Brain → All Brains Sync → Capsule Generation → Autonomous Breakthroughs |
| 02:30 | Rust (rev 75) | Phase 1-5 (enrichment) → Phase 6 (GCN 276 nodes) → Phase 7 (evidence hash) → Phase 8 (omics eval) → Phase 9 (NPEAD learning) → Phase 12 (brain feeds + capsule compression) |
Cancer Coverage
| Cancer | Heroes | Novel Targets | Validated |
|---|
| Prostate | ~70 | Standard pipeline | Multiple A+ hits |
| Pancreatic | 17 | 10 (physician-rs discovered) | 8/10 validated |
| Lung | 8 | Expanding | In progress |
| Breast | 5 | Expanding | In progress |
Key Differentiators
1. Novel target generation — CortexQ Gen v4 creates molecules for targets with zero known chemistry (SELFIES, 100% valid)
2. Closed-loop learning — System gets smarter every night without human intervention
3. Full governance — Every decision hash-chained, auditable, tamper-evident
4. 15-brain intelligence — Compressed into capsules that directly influence generation
5. No on-demand GPU — All Spot, watchdog-controlled, per-gene dedup
MindNetQ Inc. · Computational Oncology & Drug Discovery · Account: 865950114883 · Profile: edicer-pub