AI Native · Proprietary · 3+ Years R&D · Hedge-Fund Calibre
PALM
Perpetual Algo Machine
A proprietary alpha model toolkit, framework and portfolio/risk management platform — the result of 3+ years of R&D. PALM is fully systematic and mathematics-based, deploying a suite of pure alpha models across all asset classes. Not HFT, but near — operating in the medium to long-frequency trading (MFT/LFT) regime where signal persistence and execution precision both matter.

The platform's IP lies in three pillars: the AI trading agents as the decision maker and having veto rights over the mathematical models, the alpha model library (14+ distinct quantitative models spanning ML, jump diffusion, mean reversion, and momentum) and the research & analytics pipeline (an 8-step, 3–6 month development methodology from raw data to live deployment).
14+
Alpha Models
3+
Years R&D
8
Platform Layers
8
Pipeline Steps
HFT · MFT · LFT
High · Medium · Long Frequency
Three Pillars of IP
O
Pillar I — AI Trading agents
AI agents running quant models on different assets
A collection of agents bringing together analyst, researcher, risk manager, trader on top of validated quantitative models. Language models are built to process unstructured text, not to execute non-deterministic and stochastic computations resulting in trading decisions. Therefore that is done by experts, data generated are fed in to the AI models to train the results, learn from them and veto the decision making process based on multi-dimensional factors that are normally not available to human-eye.
AI Agents Automated decision-makingt Multi-Factor
📐
Pillar II — Alpha Model Library
14+ Proprietary Quantitative Alpha Models
A diverse library of alpha-generating models spanning ML-enhanced signals, stochastic jump diffusion processes, mean reversion frameworks, and multi-factor probability systems. Models are classified into four archetypes — each targeting different market dynamics and timeframes. Three proprietary sub-components (Memphis, Sensus, Rheo) serve as composable building blocks recombined across the library.
ML / AI Jump Diffusion Mean Reversion Multi-Factor
🔬
Pillar III — Research & Analytics Pipeline
8-Step · 3–6 Month Development Methodology
The systematic methodology that transforms raw data into production-grade alpha models. Each of the 8 steps is mandatory and non-negotiable. The pipeline is applicable to single assets or a portfolio within the same market. Every step combines mathematical rigour with human judgment and ML techniques — and the full cycle takes 3–6 months per asset/model combination before any live deployment occurs.
Data Acquisition Model Maths Backtest & Calibrate Go Live
Platform at a Glance
Quantitative Engine
Central computation hub. Signal processing, model execution, real-time analytics.
Data Layer
Multi-asset, multi-frequency ingestion. Tick to daily. Full normalisation pipeline.
Portfolio Management
Signal aggregation, position sizing, cross-model capital allocation by Sharpe contribution.
Risk Management
Real-time VaR, drawdown limits, position concentration controls, cross-model correlation.
Execution Management
Order routing, timing optimisation, slippage minimisation. Signal-to-fill precision per model.
Reporting
P&L attribution by model, asset, timeframe. Sharpe, drawdown, turnover transparency.
Classification
Proprietary IP
Frequency
HFT · MFT · LFT
Coverage
All Asset Classes
Approach
Systematic · Maths-Based
PALM is not publicly disclosed. This document is based on the platform description provided and is intended as an internal reference only.
Architecture
Platform Architecture
PALM is structured as 8 interconnected layers, all feeding into and drawing from a central Quantitative Analytics Engine. The architecture mirrors institutional systematic trading infrastructure — not a collection of standalone scripts, but a fully integrated platform where each layer has explicit interfaces to the others.
8-Layer Architecture
⚙️
Quantitative Analytics Engine
The central hub. All layers feed into and draw from this engine — signal processing, model execution, real-time analytics, and cross-layer coordination.
📐
Portfolio Construction & Management
Signal aggregation across all active alpha models. Position sizing via risk-adjusted contribution. Capital allocation optimised by Sharpe per model at the portfolio level.
⚡
Execution Management
Order routing and timing precision. Slippage minimisation is critical in the MFT regime (not so much on LFT regime) — transaction costs can easily eliminate the edge of a well-calibrated model if execution is poor.
🛡️
Risk Management
Real-time drawdown monitoring, VaR computation, position limits, and cross-model correlation tracking. Ensures no single model or asset class dominates total risk exposure.
🔬
Research & Analytics Pipeline
The IP-generating layer. Houses the full 7-step model development methodology — from raw data acquisition through mathematical formulation, testing, calibration, AI training, to live deployment.
🗄️
Data
Multi-asset, multi-frequency data ingestion, cleaning, normalisation, and storage. Covers tick to daily frequencies. The foundation all other layers depend on — data quality is model quality.
📊
Reporting
P&L attribution by model, asset, and timeframe. Sharpe decomposition, drawdown analytics, turnover reporting. Full transparency across all active strategies in real time.
⚡
AI agents in POD structure
Full automation by various AI agents (trading, research, quant, risk management, pod management, chief risk management) per pd generating alpha only for its edge. Several pods exist simultaneously and they are managed in real time.
🔩
Integrated, Not Modular
Unlike a collection of standalone tools, PALM's layers are explicitly integrated — outputs from the Research Pipeline feed directly into the model execution engine; risk limits computed in the Risk Management layer are enforced in real time by the Execution Management layer; Portfolio Construction draws on live Reporting outputs to rebalance allocations. This integration is itself an IP asset — it took 3+ years to build and cannot be replicated by simply assembling off-the-shelf components.
Competitive Context
Positioning
Where PALM sits in the landscape of systematic trading platforms and how its philosophy compares to elite hedge fund infrastructure.
PALM vs Elite Hedge Funds
✓
AI native. None of the existing hedge funds are natively using AI.
✓
Hedge-fund calibre pipeline rigour. The 8-step methodology with 3–6 month cycles per model matches the development intensity of top systematic shops.
✓
Math-first, ML-second philosophy. Unlike black-box ML funds, PALM grounds every model in explicit mathematical theory — jump diffusion, PDEs, stochastic processes — before applying ML enhancement and AI agents take control.
✓
Multi-asset, multi-model diversification. 14+ distinct alpha models across asset classes mirrors the pod diversification of multi-strat funds at a smaller scale.
✓
Full-stack integrated platform. Data through execution through reporting — not a signal library bolted onto a third-party OMS. Mirrors institutional infrastructure architecture.
~
Pipeline competitive with institutions & Hedge Funds — directionally. Not comparable in scale or resources, but the IP architecture and development discipline reflects the same systematic rigour.
What Makes PALM Distinctive
●
AI Architecture. Agentic architecture, once completed, will fully replicate what trading pods do (research, analyze, backtest, ingest, model, trade & manage) across different asset classes.
●
The MFT sweet spot. Not HFT (an infrastructure arms race) and not slow fundamental — PALM sits in medium frequency where mathematical edge persists long enough to capture but fast enough to avoid crowding by fundamental managers.
●
Composable internal sub-models. Memphis, Sensus, and Rheo act as building blocks recombined across the library — creating model diversity from a shared, well-understood component set. Accelerates development without sacrificing rigour.
●
Human + machine at every step. The pipeline explicitly combines mathematical formulation, ML techniques, and thorough human judgment — avoiding both pure black-box ML fragility and pure rules-based brittleness.
●
Conditional AI training. Not every model needs ML. PALM's discipline of deciding when ML adds value (vs when a clean mathematical model is sufficient) is itself a competitive advantage — it avoids overfitting by default.
Trading Frequency Spectrum


HFT
μs–seconds-minutes


MFT
minutes–hours-a few days


LFT
days–weeks


Fundamental
months–years
PALM operates in the medium to long frequency zone — fast enough that mathematical signal edges persist without requiring co-location infrastructure, but slow enough to avoid the diminishing returns of the HFT arms race. This is the regime where stochastic process models, jump diffusion, and ML-enhanced signals have their strongest edge-to-cost ratio.
Classification
Proprietary IP
Regime
HFT · MFT · LFT
R&D Investment
3+ Years
Model Library
14+ Alpha Models
PALM is not publicly disclosed. This document is based on the platform description provided and is intended as an internal reference only.
PALM · Proprietary & Confidential