
32 projects
Answer employee questions from a 300-page handbook, with citations and refusal when unsure.
Classify and route inbound support messages into validated JSON with an eval suite guarding quality.
Pull line items from scanned invoices with a vision-language model and confidence-based human review.
Gather sources, take notes and draft a cited brief, with memory that survives restarts.
Triage incidents by reading logs and proposing remediation that a human approves.
Find and fix prompt-injection and permission flaws in a deployed agent.
Predict cancellations and pick a decision threshold from the real cost of outreach.
Cluster behaviour data and validate segments against business outcomes.
Serve a ranking model behind an API and track drift.
Implement and train a decoder-only transformer and study how it scales.
Use self-supervised pretraining to reach strong accuracy with 5% of labels.
Reproduce a published result, run ablations and report what holds up.
A CI pipeline that trains, tests and only promotes models that beat production.
Monitor live predictions and alert when input distributions shift.
Route requests across models with caching and per-team budgets.
Answer questions across wikis and code with citations and freshness.
Ensure users only retrieve documents they can open.
Flag manufacturing defects with a confidence threshold tuned for recall.
Track and count vehicles from video with an events dashboard.
Architecture for routing, caching, quotas and audit across teams.
A multi-year platform design with controls mapped to governance requirements.
Decide whether a campaign changed conversion, with the right test and caveats.
Predict default risk and explain drivers to a credit committee.
Forecast weekly product demand with backtested accuracy.
Estimate the causal effect of a price change using observational sales data.
Turn demand forecasts into an optimal rota under labour constraints.
Allocate offers adaptively while measuring lift.
Model, load and test an analytics warehouse with documentation.
Sessionise hundreds of millions of events with Spark.
Stream orders into a lakehouse with live dashboards.
A secured Power BI report with drill-downs by region and rep.
Find where and why users drop out of a sign-up funnel.