Building AI systems where scientific reasoning, quantitative analysis and real-world decisions intersect.
At Imperial College London biochemistry meant protein structures, cryo-EM, and NMR. Now at IE University in Madrid I build AI systems — for AstraZeneca oncology, Amazon logistics, weather forecasting, and training intelligence. The common thread: quantitative depth and scientific literacy applied to problems that require both.
AI deal desk for Amazon Shipping Enterprise BD. Ingests any mix of RFQs, CRM notes, emails, and spreadsheets into a canonical Opportunity Profile. Deterministic serviceability engine maps requirements against a capability matrix; pricing engine produces 3 margin scenarios with finance guardrails; logistic win-probability model trained on 360 historical deals. Generates executive summary, risk assessment, commercial strategy, and a branded client-ready PPTX proposal.
AI camera coach for goal-based training. In-browser pose estimation (MediaPipe, 33-point full-body model) with rep counting, form scoring, ROM/tempo/time-under-tension tracking, and per-rep live cues. AI plan generator builds week-by-week programmes. 115-exercise library across tier-1 full camera coaching, tier-2 rep counting, and tier-3 manual logging. Session logging and insights dashboard; AI coach grounded in training history.
AI-governed RAG platform for AstraZeneca Spain's oncology HCP engagement. Classifies every query by topic, intent, risk, and compliance route before retrieval, routing across 7 compliance pathways; generates demand intelligence from every interaction. Scored 4.9 vs 3.6 vs AZ Engage Copilot on weighted selection matrix. Financial model: €93K Y1 net investment, cash-flow positive Y2, €1.1M cumulative 3-year FCF.
Geospatial crisis navigation platform for civilians in conflict zones, built in 24 hours. 3D globe dashboard with 3 live data layers — conflict (ACLED), flight (OpenSky), and seismic (USGS) — AI crisis assistant powered by Groq, and crowdsourced hazard reporting via Supabase with PostGIS.
Physics-informed ML post-processing of NWP forecasts to correct systematic local errors for Madrid and central Spain. Ingests 3 NWP model sources (IFS, AIFS, ICON-EU), ECMWF Open Data, and AEMET observations into DuckDB. Custom meteorological feature engineering; LightGBM bias-correction layer with SHAP explainability; 39 unit tests. FastAPI backend; Next.js dashboard.
Seq2seq transformer for bi-directional translation between SMILES strings and IUPAC chemical names, trained on 500K+ PubChem compound pairs with custom tokenisers. PR #4931 merged to DeepChem.
Scheduled agentic pipelines delivering daily pharma and biotech intelligence briefs. Covers oncology, peptide/GLP-1, neuro/CNS, and AI drug discovery. Draws from 5 live data sources: FDA.gov, SEC EDGAR, ClinicalTrials.gov, PubMed, bioRxiv. Separate pipeline tracks FDA catalyst calendars with PDUFA dates, AdCom votes, and binary event classification. Running daily.
Portfolio-aware market intelligence platform delivering morning briefs, pre-open intraday setup, and event-driven breaking alerts. Monitors holdings, screens macroeconomic signals, and pushes summaries via Telegram and email on a live schedule.
Personal health intelligence platform synthesising wearable biometrics, manual check-ins, and training logs into a deterministic daily readiness score and training mode. Local-first v0.1, with planned Garmin and Oura integrations and N-of-1 supplement timing experiments.
IE Business School + School of Science & Technology, Madrid
Imperial College London
The Manchester Grammar School
Chemistry (A*), Mathematics (A*), Biology (A), German (A). Piano Grade 8, Violin Grade 8.
Open to roles in pharma tech, fintech, biotech, and data analytics. Particularly in Spain. Spain and UK citizenship.
jacksangster.033@gmail.com