Index
Papers
Working systems and research, written up as preprints. The decisions and trade-offs, not the highlight reel.
Lineup
J. Cousins
A production data platform for live-music discovery. 1M+ events ingested daily from six APIs; entity resolution via a persistent cross-source identifier registry; SwiftUI iOS client. Twelve months of production operation; zero data loss.
NutriPlan
J. Cousins
Personalised meal planning by vector retrieval over a curated database with subsequent LLM optimisation against macro targets. Calorie accuracy ±10%; sub-2s generation; 99.8% backend uptime.
JobHunter
J. Cousins
Encoding personal career goals as an automated job-discovery and scoring system. Six-dimension framework evaluated by Claude API; daily crawls across four sources; Notion as triage UI.
Earnings-call Q&A alignment as an equity signal
J. Cousins
A research replication of Chiang et al. (2025): contrastive PyTorch classifier over FinBERT embeddings of earnings-call Q&A pairs, evaluated against forward equity returns. Architecture and dashboard complete; labelled training set is the critical-path blocker.
A personal AI layer
J. Cousins
OpenClaw (a self-hosted local agent surfacing through Telegram) and a Claude Cowork-backed assistant: two tools sharing one job description, deliberately not unified for cost / latency / reasoning reasons.
An adaptive learning system
J. Cousins
A self-directed curriculum across seven technical subjects, with spaced testing and Claude-mediated curriculum updates. Treats the syllabus itself as state an LLM can read, write, and reason about.
SMB AI discovery
J. Cousins
A structured discovery method for identifying real agentic-AI use cases inside small and mid-sized businesses, adapted from enterprise discovery work in a €176B European asset manager.
See also
Past work and selected coursework, archived separately.
The UCL Physics dissertation, the DataCamp Data Science certification project, and other archived material live on the past-work page.