AI that remembers
what worked, and when.
HumAIn Initiative builds vDNA persistent domain memory for inference architectures. NEXUS validates the approach in live quant trading today.
The architecture that makes memory possible
Three memory strands work in parallel. Trade Memory captures what happened. Conversation Memory captures why. Meta-Learning adjusts what fires next.
Trade Memory
Every executed trade writes a structured memory strand — regime, confidence, and outcome preserved for adaptive replay.
Unique trade fingerprint
Macro context at entry
Confidence score at signal time
Triggering signal pattern
Realised return, post-close
What the model updated post-trade
Conversation Memory
Analyst intent and decision rationale persist across sessions. The model recalls prior reasoning, not just prior positions.
Session scope boundary
Declared objective this session
Topic the analyst probed
Response fingerprint for replay
What changed after this session
Parameter delta recorded
Meta-Learning
Across all strands, the meta-layer tracks which strategies work in which regimes and adjusts weighting before the next trade fires.
Encoded market state
Current regime classification
Aggregate confidence this regime
Weight applied to this strategy class
What the meta-layer changed
Evidence for the weight change
Each vDNA strand is a persistent domain memory record written at trade close and queryable in real time.
NEXUS validates the architecture in real markets today.
Every trade NEXUS executes writes a structured memory entry: regime tag, confidence score, outcome, and an adaptation note the meta-learner reads on the next decision. The architecture is running, not theoretical.
Real trades. Real memory. Real adaptation.
NEXUS writes a vDNA memory entry on every closed position. The system learns which regime tags predicted which outcomes, and shifts strategy weights accordingly without retraining.
Persistent memory
Trade history accumulates across sessions. NEXUS does not forget a regime pattern when you close the terminal.
Regime-aware confidence
Each decision carries a strategy confidence score derived from outcomes in the same tagged regime, not global averages.
Adaptive weighting
After each closed trade, the meta-learner adjusts strategy weights. No manual tuning, no batch retraining required.
Nov 14, 2025
09:31:04 UTC
symbol
ESLONG
market_regime
strategy_confidence
pnl_bps
+34 bps
adaptation_notes
High-volume open confirmed regime. Confidence weight raised 0.04 for similar pre-market conditions.
Nov 14, 2025
10:47:22 UTC
symbol
NQSHORT
market_regime
strategy_confidence
pnl_bps
-18 bps
adaptation_notes
Regime mis-tag at open. Meta-learner flagged correlation with CPI release day. Strategy weight adjusted down 0.06.
Nov 14, 2025
13:02:55 UTC
symbol
ESLONG
market_regime
strategy_confidence
pnl_bps
+51 bps
adaptation_notes
Post-midday trend confirmation. Highest confidence score in session. Memory strand updated with regime sequence.
Illustrative entries. Field structure mirrors the live vDNA schema running in NEXUS today.
Architecture efficiency
The model gets smarter without the cost of retraining.
vDNA memory stores domain knowledge as structured records, not weight updates. These numbers reflect the architectural expectation — each is under validation in live NEXUS deployments.
Parameter reduction
A vDNA memory strand encodes domain context in structured records rather than billions of floating-point weights — the base model shrinks dramatically and runs on commodity hardware.
Energy savings vs. full retraining
Writing a new memory strand costs micro-seconds of compute. No GPU cluster, no multi-day training run — the model adapts through structured records, not gradient descent.
Marginal training cost per new context
Each trade outcome appends one vDNA record. The architecture was designed so adaptation never requires a retrain — the cost curve stays flat as the strategy universe grows.
All figures are architectural projections under active validation in NEXUS Pro deployments. They do not constitute financial or performance guarantees.
vDNA architecture
Three strands. One persistent memory.
Each strand captures a different layer of context. Together, they give the model a typed, queryable record of every market condition, every analyst decision, and every structural lesson learned.
What this strand stores
market_regime
The macro context active at trade time: trend, mean-reversion, or volatility compression.
strategy_confidence
A scored estimate of how well the strategy matched conditions at entry.
outcome
Raw PnL, risk-adjusted return, and whether the regime call was correct.
adaptation_notes
Structured observations the system writes back after every closed position.
Adaptive weighting
Adaptive weighting re-scores each historical trade entry against current conditions. Trades executed in a matching regime carry higher influence; distant-regime trades decay in weight over time without deletion.
What it enables
The system knows which strategies have an edge in today's market structure, because it has a typed memory of every regime it has navigated and what worked inside each one.
Persistent across sessions. No retraining required.
Strategic Pilot Partnership
$5,000
for a 90-day engagement
One firm, one slot, founder-led. We embed into your workflow for 90 days to document, measure, and validate the vDNA architecture against real trading data.
What the engagement includes
Architecture documentation
Full vDNA schema specification, integration blueprints, and annotated sequence diagrams tailored to your stack.
Co-developed benchmark methodology
We build the measurement framework together. You own the results; we publish the methodology as a joint whitepaper.
Hosted integration support
Founder-level engineering time throughout the 90 days. Direct async channel, weekly sync, and escalation SLA.
Joint case study rights
Both parties retain publication rights. You can cite benchmark results in investor decks; we can cite them in product collateral.
Inquire about a pilot partnership
Performance figures cited during the engagement are under validation. Nothing in this pilot constitutes financial advice or a guarantee of trading outcomes.
Common questions
One slot open now.
The engagement is founder-led. We run one at a time, with complete focus. Submit an inquiry and we will respond within two business days.
Common questions
Answers, directly.
Everything you need to evaluate vDNA architecture and NEXUS Pro before reaching out.
Still have a specific technical question? Email the founders directly.
Get started today
Two ways in.
NEXUS Pro gives individual quant traders a live vDNA memory layer starting at $399. The Strategic Pilot gives firms a 90-day co-developed integration.
NEXUS Pro early access
A hosted vDNA memory layer for individual quant traders and small algo trading firms. Request access now to join the pilot cohort.
Strategic Pilot Partnership
A 90-day white-glove engagement for one trading firm or cloud platform operator. Covers architecture documentation, co-developed benchmark methodology, hosted integration support, and joint case study rights.
- Architecture documentation tailored to your stack
- Co-developed benchmark methodology
- Hosted integration support throughout
- Joint case study rights, NDA included
- Priced to cover infrastructure, founder time, and legal review
One active pilot partnership at a time. Reach out to check availability.