Tested separately.
Each rule is assessed for its own marginal contribution—not only the system it happens to sit inside.
BEAK TECHNOLOGIES
Quantitative research.
Institutional technology.
Proprietary models. Integrated execution. Embedded risk controls. Research and engineering, developed in house.
Explore our technologyTHE EDGE / SHORT-HORIZON RESEARCH
An edge measured
in seconds.
Beak develops quantitative technology for short-horizon market opportunities. Its research combines price-cycle modelling with the execution engineering required to act inside a finite decision window.
Price-cycle signals can decay within seconds. The research question is inseparable from the engineering question: can a candidate still be valid when the decision reaches the market?
Latency matters. Liquidity matters. Our own flow matters.
SIGNAL HALF-LIFE / EDGE DECAY
Conceptual signal response against latency. Not a performance claim or a promise of zero market impact.
THE FIRM / OWNED TECHNOLOGY
The engine
is the product.
Beak owns the research, modelling, execution and risk technology in house. This integrated stack allows models and their operating controls to be engineered together—from hypothesis to institutional deployment.
Structure within the signal.
Proprietary models translate market observations into a normalised, bounded output. A candidate still has to clear every required control.
EXTERNAL INPUTMarket data feeds
BEAK-OWNEDResearch · Modelling · Execution · Risk
EXTERNAL ACCESSTrading venue
Instrument-agnostic by design.
Current deployment scope: exchange-traded futures and foreign exchange. Historical evidence originated in digital-asset markets and does not establish performance in these deployment markets.
RESEARCH PROVENANCE / MATHEMATICS TO MARKET
Mathematics.
Engineered for markets.
04
YEARS OF MODEL DEVELOPMENT
Built in house. Tested, challenged and continually refined.
Four years of model development bring six technical disciplines into one research process. The work continues as markets change: finding structure, challenging the model and engineering its decision path.
BEAK / RESEARCH WORKSPACESIX CONNECTED DISCIPLINES
DISCIPLINE / 01
Stochastic processes
Markets are modelled as noisy processes with structure. A claim about a pattern is a claim about a distribution.
Research informs the model. The model remains subject to validation and every execution control.
12yearsTECHNOLOGY LEAD
Low-latency trading systems & broker infrastructure.
14yearsTRADING & INVESTMENTS LEAD
Proprietary trading & quantitative research.
THE ENGINE / CONTINUOUS DECISION PATH
From market data
to controlled execution.
The engine continuously reads, scores and checks the market. Research, model inference and execution controls operate as one decision path—with no human intervention in live execution and a record of every decision.
Raw observations become structured inputs for model evaluation.
Conceptual illustration, not production architecture.- 01
Read the market
Order books, transactions and liquidity from the instruments in scope enter the engine.
- 02
Score the setup
More than 100 inputs from five independent data families resolve into one state vector.
- 03
Ask the model
An ensemble evaluates the state. It returns confidence, not permission to trade.
- 04
Run eight checks
Every check must pass in the same decision window. One failure cancels the candidate.
- 05
Enter and exit
The automated position lifecycle completes inside sixty seconds. Each decision is logged.
CORE TECHNOLOGY / NORMALISED CYCLE MODEL
Different instruments.
A consistent model language.
The oscillator is normalised and bounded between −1 and +1. A reading such as +0.8 is expressed on the same scale in a volatile period and a quiet one. That is a design property—not proof of equal performance across markets.
Signal amplitude
Adjust the amplitude. The raw signal changes; its normalised counterpart remains on the same bounded scale.
01 / RAW INPUTPeak magnitude 1.0
Relative units · Fixed axis ±21.0× reference
02 / NORMALISED OUTPUTBound ±1
Dimensionless · Fixed axis ±1Same observation window
RAW SIGNALDIVIDE BY PEAK MAGNITUDEBOUNDED OUTPUT
Simplified teaching example—not Beak’s production algorithm, market data or performance. Normalisation changes the scale; it does not establish predictive value.
Comparable readingsNormalisation reduces the dependence on raw amplitude when interpreting model state.
Cycle-aware modellingThe prevailing cycle length informs the model rather than a fixed market tempo being assumed.
Candidates, not ordersA cycle turn starts the decision process. The downstream controls exist to reject candidates.
SIGNAL FABRIC / ENSEMBLE INTELLIGENCE
100+ inputs.
One coherent market state.
More than 100 inputs from five independent data families describe the market state. Agreement across competing model families informs confidence; one model agreeing with itself is not independent confirmation.
FEATURE CONTRIBUTIONS / STATE CONVERGENCE
ENSEMBLE / SCHEMATIC VOTE
M₁
M₂
M₃
M₄
M₅
Model agreement → Confidence → Eight required checks
100+inputs per scan
5deployed data families
±1normalised output
Historical digital-asset research used six families: price, order book, derivatives positioning, on-chain flow, cross-asset context and sentiment. On-chain flow is market-specific; the deployed configuration uses five families.
EXECUTION / SHORT-HORIZON ENGINEERING
Capture a move.
Do not hold a view.
The automated position lifecycle completes inside sixty seconds. Entry, predefined risk, target and exit belong to one engineered sequence—not an overnight thesis.
<60seconds / position lifecycle
- 01Predefine stop
- 02Qualify entry
- 03Manage position
- 04Exit & log
RESEARCH DISCIPLINE / WALK-FORWARD VALIDATION
Research that earns
its place in the engine.
AI-assisted research expands the number of models Beak can examine. Admission to the engine demands more: unseen-data testing, measured marginal contribution and explicit accounting for every variant tried.
Discarded when weak.
A rule that cannot be distinguished from luck is removed, however attractive its historical curve appears.
Counted honestly.
The number of variants tried remains part of the evidence. A broad search can inflate the apparent winner.
No future information.
Each walk-forward fold trains on the past and tests on unseen observations. Information must not leak backwards across the boundary.
Blue: training window Gold: unseen data
IN-HOUSE COMPUTE
GPU-supported research enables more model variants to be examined. Faster testing does not replace validation discipline.
FROM HYPOTHESIS TO DEPLOYMENT
Six connected stages- 01
Research
Define a market hypothesis and a measurable question.
- 02
Model
Build and compare competing explanations.
- 03
Validate
Challenge assumptions against unseen data and stress conditions.
- 04
Engineer
Integrate execution, pre-trade checks and system monitoring.
- 05
Productise
Document, harden and prepare the technology.
- 06
Institutional deployment
Integrate within defined technical and operational boundaries.
PRE-TRADE CONTROL / EIGHT REQUIRED CHECKS
Risk control
precedes every order.
All eight pre-trade checks must pass inside the decision window. A single failure cancels the candidate. The engine fails closed and defaults to flat.
EIGHT CHECKS / INTERACTIVE STUDY
07 / 08CHECKS PASSING
Choose any check to fail. One is enough.
CANDIDATE WITHHELD
Model confidence fails. One failed check is sufficient to withhold the candidate.
Synthetic demonstration. No live orders or production thresholds. All eight checks must pass within the decision window—not a majority vote.
01Spread and slippage budget
Execution costs must remain inside the permitted budget.
02Book depth imbalance
The available order book must support the candidate.
03Model confidence
The model must clear its confidence threshold.
04Cycle coherence
The candidate must be consistent with the cycle state.
05Volatility regime
Current volatility must be within the permitted regime.
06Per-instrument exposure cap
Instrument-level exposure must remain within its limit.
07Loss-streak state
The engine must not be in a loss-triggered lockout.
08Latency budget
The decision must complete within its latency budget.
Stand-down is part of the system.
Losses trigger a defined sequence—not a discretionary response.
- 01 / HALT30 min
Three consecutive losing trades trigger an automatic halt.
- 02 / EXTEND1 hour
If the streak continues, the lockout extends and the model retrains on the most recent window.
- 03 / STAND DOWN24 hours
If the streak persists, the engine treats the regime as hostile and stands down.
- 04 / RE-ARMThreshold-led
Retraining and threshold clearance are required. Elapsed time alone is insufficient.
Lockouts cannot be shortened manually. Stops cannot be widened mid-trade. Exposure contracts as volatility rises.
LOSS PROTOCOL / COMPILED PROTECTION
Protection that
cannot be overridden.
A losing streak may indicate that the regime has changed. The engine reduces activity first, then re-examines the model. Its return is governed by controls, not conviction.
STATE MACHINE / HALT → RETRAIN → RE-ARM
CONTROL LAYER / PREDEFINED STOPS
Every trade carries its stop before it exists.
The stop is part of the entry decision and sized to the volatility regime. It cannot be widened once the position is open.
EVIDENCE / CLEAR BOUNDARIES
A disciplined path
to deployment.
Validation methodology, control architecture and live incubation form the basis for examining the technology. Each stage carries its own evidence standard—and its own limitations.
01 / HISTORICAL EVIDENCE
Digital-asset markets
The historical, backtested and live evidence to date originated here. Backtests are not live results; evidence remains unaudited where applicable.
02 / DEPLOYMENT SCOPE
Futures & foreign exchange
The current deployment scope is exchange-traded futures and foreign exchange. Evidence from digital assets does not establish performance in either market.
03 / VALIDATION STANDARD
Method before promotion
Walk-forward testing, variant accounting, control architecture and live incubation provide the basis for examination. Illustrative charts are not an investment record.
A monthly aggregate can hide individual losses and path dependency. Risk controls reduce risk; they do not eliminate losing days, losing periods or the possibility of loss.
HONEST LIMITS
Designed around
real market constraints.
Short holding periods and automated controls do not remove market or infrastructure risk.
Directionless markets
When no clean cycle is present, usable edge can diminish. The appropriate output may be less activity or no candidate.
Sudden gap moves
A price jump can skip an intended exit level. A short horizon reduces time exposed, but cannot make every price tradable.
Venue disruption
Outages, trading halts and liquidity collapse can prevent execution as designed. External venue dependence remains a technical constraint.
Crowding & market impact
More flow pursuing the same short-horizon opportunity can compress edge and degrade fills. Capacity cannot be assumed to be unlimited.
OPERATIONS / OWNERSHIP & ASSURANCE
Owned technology.
Defined accountability.
In-house GPU compute supports research and model testing.
Conceptual illustration, not a photograph of Beak equipment.COMPONENTRESPONSIBILITY
Market data feedsExternal
Research & modellingBeak-owned
Execution & risk engineBeak-owned
Trading venueExternal
Decision recordsLogged internally
Decision-level traceability.
Internal decision records support investigation and reconciliation. They are not a substitute for independent verification.
A defined technology mandate.
Beak is not presented as a custodian, fund or investment manager. Technology ownership and external market access are distinct responsibilities.
People and system continuity.
Automated execution and logged decisions support operational continuity. They do not replace the specialist people responsible for ongoing research.
ASSESSMENT STANDARD
No independently audited track record, independent administration or multi-venue redundancy is asserted by this website. Such capabilities require separate, current evidence.
THE ENVIRONMENT / QUANTITATIVE RESEARCH TERMINAL
The system,
in view.
Explore the research terminal: cycle models, factor scores, indicator families and control gates, recalculated together from synthetic market data.
SYNTHETIC MARKET REPLAY
No live data. No live orders. Not investment performance.
ResearchMulti-factor inputs and normalised cycle models.
ExecutionCandidate decisions and the eight-check control path.
System healthIllustrative telemetry and risk-control states.
CAPACITY & LIMITATIONS
Finite edge.
Finite capacity.
Market depth, execution costs and market impact constrain deployment. Beyond a finite threshold, a strategy’s own flow can degrade fills and compress edge.
The curve illustrates the relationship only. It is not a measured capacity result or an investment projection.
Founding-team modelling. Unaudited. Not demonstrated at this size.
Founding-team modelling. Unaudited. Not demonstrated at this size.
These figures describe modelled limits—not available investment capacity, proven scalability or guaranteed results. Market impact must be examined in deployment.
THE TEAM / LEADERSHIP & OPERATIONS
Four people.
One shared discipline.
Company leadership, technology infrastructure and operational support work together. The founding team runs trading models and research collectively under the Chief Executive Officer.
01 / COMPANY LEADERSHIP
Jonathan Low
Founder, Director, Chairman & Chief Executive Officer
Leads business development, partnerships and origination, connecting Beak’s research and technology with institutional relationships.
Ampyre Group · BipTap · 1TX0
02 / Management & support
Johanna Boom
Shareholder, Management & Backend Support
Leads operations, bringing five years of management experience in the banking and finance sector.
03 / Technology infrastructure
Vinoth Phongpasirth
Chief Technology Officer
Responsible for technology infrastructure, system operations and vendor management.
04 / Local operations
Somluoy Thorlavong
Operations Lead, Laos
Leads administration, statutory filings, banking and regulatory liaison, and accounts in Laos.
CONTINUITY / BY DESIGN
Built into the system.
Not dependent on one individual.
The engine runs without a person in the live execution loop. Every decision and fill is logged. Risk controls sit in the execution path, and research is documented as a repeatable procedure.
People remain responsible for research, infrastructure and operations. This system design supports continuity; it does not eliminate operational or market risk.
Technology, with clear boundaries.
Beak Technologies is a quantitative research and technology company—not a fund or investment manager. Research, modelling, execution and risk technology are owned in house; market data feeds and trading venues remain external.
INSTITUTIONAL ENQUIRIES / TECHNICAL WALKTHROUGH
Let’s examine
the technology.
hello@beak.tech For institutional, government and research enquiries, contact the team to discuss the architecture, control framework and deployment scope.
Request a technical briefingResearch & engineering careers
careers@beak.tech
This website describes research and technology. It is not an offer of an investment product, investment advice, or a guarantee of results. Illustrative visualisations and synthetic demonstrations are not performance evidence.
Design summary, not a specification.System diagrams explain the approach for technical readers. They are illustrative, subject to change and do not expose proprietary production parameters.
Controls are not guarantees.Automated trading technology remains exposed to market, execution and infrastructure risk. Short holding periods, stops and lockouts do not eliminate loss.
Research detail under agreement.A technical walkthrough can cover the engine, signal families, research methodology and control logic. Detailed supporting material is discussed separately with the team.