Know what the board can decide—and what the pack cannot prove.

Turns clearly labelled meeting notes into decisions, actions and open questions, with the exact source line kept beside every item.

• Browser-local usable workflow
• LangGraph human pause and resume
• LangSmith-ready synthetic evaluations

Pick England's matchday mood, pressure response, and wildcard player. The app runs 10,000 transparent World Cup scenarios and gives you a result you can share.

• Seeded Monte Carlo simulation
• Shareable URL scenarios
• Visible assumptions and model range

London event data is scattered across venue pages. This private pipeline gives each source explicit config, extraction rules, dedupe, checks, and scheduled runs.

• 103+ source configs
• Playwright crawling
• 167-test reliability suite

Upload notes or documents and turn them into flashcards, quizzes, and a spaced-review plan. Source material stays part of the review loop.

• PDF/DOCX/TXT ingestion
• Background generation jobs
• SM-2 review loop

Normalises noisy job listings into salary, skills, remote-work, and hiring-volume signals for a weekly market brief.

• Listing ingestion and normalisation
• Salary and skill trend reports
• Alerts and API tier design

Moves same-day cover from school request to agency assignment and teacher offer, with eligibility and live status in one workflow.

• Sequential assignment engine
• School, teacher, and agency portals
• Server-sent event updates

Keeps a resident request's status, owner, evidence, and payment context together from intake to triage.

• Resident request intake
• Operations console and audit trail
• Payment and AI triage concepts

Turns campaign CSVs into DuckDB tables, model features, checks, and Streamlit views that rebuild from demo data.

• Bronze/silver/gold flow
• XGBoost model training
• Streamlit dashboard

Reconciles board packs, RAID, actions, decisions and schedule evidence before the board, then records the human outcome and open conditions.

• Board-readiness review across five evidence groups
• Source-linked conflicts, gaps and decision questions
• Approval-condition follow-up

Retrieves similar project changes, explains why each case matters, and shows the intervention and measured outcome before another decision is approved.

• 16 synthetic precedent cases
• Inspectable retrieval and evidence graph
• Human review and outcome feedback loop

Packages dataset preparation, conditional GAN training, generated outputs, and a small API path.

• Conditional GAN
• Dataset preparation
• API integration

Uses swipe history to rank unseen profiles, with temporal holdouts and Top-K metrics instead of a random train-test split.

• Implicit-feedback ranking
• Temporal holdout
• Top-K metrics

Packages HR, absence, and sales data into a stakeholder review pack with PBIX files, prepared data, screenshots, and commentary.

• Summary, sales, and absence views
• Prepared CSVs and PBIX
• Stakeholder-ready writeup

Compares embedding models for retrieval and shows where semantic similarity stops being evidence.

• Transformer embeddings
• Cosine ranking
• Retrieval caveats