thinkingML
Research
From foundational methodology to production systems
Research driven by production reality
Problems become methodologies. Methodologies become systems.
Our research emerges from production constraints: optimising models under computational limits, adapting systems to shifting data distributions, building governance that survives regulatory examination.
Systems become competitive advantage
Applied Projects
Unstructured Document Data Extraction
Production pipeline processing over one million complex financial documents weekly. Extracts structured data from unstructured sources using adaptive parsing, entity recognition, and validation algorithms replacing brittle legacy parsers with Agentic AI.
Cross-Asset Trade Surveillance
Real-time anomaly detection system monitoring trading activity across multiple asset classes. Identifies patterns indicative of market abuse, insider trading, and regulatory violations. Designed for high-throughput environments with low-latency precision where standard rule-based systems fail.
Deep AI Analytics for Portfolio Management
Predictive analytics platform for position and portfolio risk assessment. Combines market data, position data, and predictive models to provide real-time risk exposure analysis, stress testing, and scenario modelling for institutional portfolios.
Agentic RAG Systems
Retrieval-augmented generation systems with agentic decision-making for complex query resolution. Combines vector search, knowledge graphs, and LLM reasoning to provide contextually accurate responses across large document repositories with audit trails for compliance.
Core Research
Evolutionary Cellular Optimisation (ECO)
A generative optimisation algorithm that constructs its search space during execution. By evolving hyperparameters as dynamic cellular structures, ECO treats optimisation as discovery, not selection, building viable configurations through generative constraint rather than sampling from fixed candidate sets.
LoRA with LLaMA-3 on FreshQA
Lightweight adaptation of LLaMA-3 8B using low-rank adaptation (LoRA), guided by ECO for dynamic hyperparameter control. Evaluated on unseen QA data with deterministic metrics under 8×H200 conditions.
Medical Image Processing (XRAY/ROCT)
Unified framework for medical image optimisation on retinal and chest datasets. Supports structured benchmarking across model classes, data types, and optimiser settings. Noise conditions are fully configurable for environment-specific tuning. Foundation of PhD research in robust perception under degraded conditions.
Cross-Domain Optimisation Evaluation
Stress-testing generalisation across divergent landscapes. Evaluates ECO's resilience by interleaving medical imaging, structured and unstructured text data and reporting summaries. Demonstrates stability in high-noise environments where traditional optimisers collapse.