Research Ledger
A structured archive of applied research across mathematics, markets, systems, education, and technology. Each record begins with the question behind the work, then documents the method and the resulting system.
Index of records
21 active records
How can a learning system infer a student's changing knowledge state and select the next problem that maximizes useful progress?
Adaptive AI training platform that models a student's knowledge over time, selects the next useful problem, and evaluates written reasoning—not just final answers.
How can fragmented competition information be transformed into a reliable recommendation path for each student's interests and level?
AI navigator for UK and international academic competitions—cataloguing opportunities, tracking deadlines, and recommending next challenges from a verified catalogue.
How can weak behavioural and on-chain signals be combined to distinguish coordinated identities from legitimate applicants?
Crypto fraud detection layer combining login/signup behaviour, allocation patterns, and on-chain wallet evidence.
How can public account quality be measured transparently without collapsing heterogeneous evidence into an opaque score?
Full-stack prototype that scores public X accounts on a transparent 0–100 scale from official API data, with categorized evidence and a shareable report UI.
How should reputation be distributed and updated when user activity follows a heavy-tailed rather than uniform distribution?
Phase-1 audit and scoring prototype for a crypto social platform: heavy-tail distribution analysis, rarity-driven achievement automation, and no-downgrade rollout simulation across 24k+ users.
How can hierarchical financial disclosures be reconstructed reliably when their structure is encoded visually across irregular tables?
PDF extraction pipeline for hierarchical geography and segmental disclosures in financial statements—detecting multi-level table headers and normalizing nested breakdowns into auditable structured feeds.
How can instrument identities be inferred across disconnected systems when no complete common identifier exists?
Research-style reconciliation framework that uses trade logs to infer instrument-ID mappings across disconnected financial systems.
How accurately can disclosure timing be forecast from historical release behaviour and issuer-specific reporting patterns?
Latency-sensitive earnings-release intelligence for ~2,000 bond issuers—forecasting disclosure windows, monitoring issuer pages, and capturing PDFs before standard feeds.
How can investor conviction be estimated from post-distribution token behaviour across wallets, staking, and liquidity positions?
Post-IPO / ICO investor conviction score measuring token retention after distribution, including wallet balances, staking deposits, and LP exposure.
How can unstructured debt disclosures be linked to canonical bond records with an auditable level of confidence?
PDF debt-note extraction and Markit bond matching—linking ISIN-level disclosures in filings to issuer reference data and quarterly financials.
How can model lineage, runtime behaviour, and parameter sensitivity reveal whether complex reinsurance runs remain trustworthy?
Operational dashboard and lineage layer for multi-model reinsurance runs on Tyche—tracking outputs, runtime, treaty KPIs, and parameter sensitivity.
How can execution quality be classified fairly when bond trades must be compared against several incomplete market benchmarks?
Hedge fund compliance system for labeling bond execution quality across Bloomberg, Markit, dealer quote, and trade-print benchmarks.
How can semi-structured human trade messages be converted into deterministic events without losing ambiguity or provenance?
Hedge fund trade-capture parser that converts semi-structured Bloomberg IB messages into normalized trade events across multiple asset classes.
Which combination of anomaly detection and supervised learning best ranks rare fraudulent transactions for constrained human review?
Explored anomaly and supervised models on a severely imbalanced card-transaction dataset to rank transactions for analyst review.
How can stateful FIX messages be normalized into reliable market events while preserving ordering, validation, and traceability?
Hedge fund execution pipeline that normalizes FIX messages into validated market events for analytics and controls.
How can bond and CDS exposures be sized so that spread views are isolated from first-order interest-rate risk?
Credit markets hedging tool that balances long bond exposure with CDS positions using DV01-driven sizing.
How much explanatory value do trait rarity and liquidity contribute to the relative pricing of unique digital assets?
Interpretable NFT marketplace screener combining trait rarity, listing prices, trait floors, and liquidity confidence.
How can overlapping spatial entities be arranged to preserve topology while minimizing visual collision and distortion?
Spatial systems optimizer for overlapping entity nodes, using planar ellipse layout and topology constraints.
How can candidate fit be ranked across skills, experience, environment, and constraints while keeping every recommendation explainable?
Interpretable candidate ranking that combines hard-skill fit, environment fit, experience, and constraints—no neural black box.
How can a person's likely profession be inferred from the structure and balance of their observed hard skills?
Hard-skill extraction from CVs, skill-relationship graph, and profession inference via balanced vector matching.
Can longitudinal professional behaviour estimate compatibility with a working environment more reliably than static self-reporting?
Observable behavioural signals from professional platforms used to estimate candidate–environment compatibility over time.