CV

• Build Power BI dashboards that track classification adoption and compliance across more than 10,000 users, so that leadership can see risk exposure and adoption trends.
• Query Microsoft 365 audit logs and Purview data with SQL and Python to analyse user behaviour, sharing patterns and how well policies are working.
• Automated the weekly reporting workflow in Python, which cut the time spent on manual reporting by about 70%.
• Map how data is ingested, validated and handled when something goes wrong, and define controls such as record counts, checksums, audit logging and reconciliation so that data issues stay traceable and testable.
• Write requirements, data definitions, user stories and acceptance criteria in Azure DevOps, and interview business owners and engineers until both groups agree on one view of the solution.
• Support a global Microsoft 365 data-protection rollout reaching more than 10,000 end users, including training and follow-up.

• Interviewed technical teams at more than 100 client companies and wrote the technical narratives for their HMRC R&D tax relief claims, covering software, data and engineering projects.
• Reconciled project cost datasets in SQL and Excel to find anomalies before submission.
• Turned complex technical detail into clear, evidence-based reports for tax specialists, written to HMRC's criteria for qualifying R&D.
• Redesigned a client's ETL process so that it no longer collected duplicate form fields or extracted data nobody used, and presented the impact to the client.
• Looked into why a client was making so little use of their Salesforce CRM, compared how they worked with how they could work, and recommended changes, delivered in Scrum sprints with regular client reviews.

• Prepared and reconciled client financial data for US and UK tax filings, checking that records were complete and consistent before statutory deadlines.
• Ran checks for missing or inconsistent information and explained what I found directly to clients.

• Wrote a Python package (NumPy, ASE, LAMMPS) that generates random polyethylene structures and relaxes them with a force field, following the AIRSS structure-search method.
• Presented the results as a poster at the In2Research summer conference.

• Built an end-to-end pipeline loading 7.1 million HM Land Registry sales (2018–2024) and ONS earnings data into PostgreSQL through raw, staging, star-schema and mart layers, feeding a four-page Power BI report.
• Reconciled every row (raw always equals kept plus rejected, with 1.47 million rejections each given a reason); council medians match the official ONS figures within 2% for 99.7% of council-years.
• Mapped 46 renamed or merged districts to current councils in a documented seed file, and aligned every sale to ONS October–September years; unit tests run on GitHub Actions.
• Found that 111 of 294 councils became less affordable between 2019 and 2024, while London improved most (median ratio 12.9 to 12.0; Mann-Whitney U, p < 0.000001). Code on GitHub

• Built a Python pipeline to extract match data from the Riot Games API within a limit of 100 requests per 2 minutes, using throttling and staged JSON caching so runs could stop and resume.
• Requested about 3,000 matches, removed duplicates to leave 1,847 unique games, and loaded them into MySQL alongside a public 9,900-game 2020 dataset.
• Aggregated player-level data into team-level features to match the 2020 schema, and used a correlation matrix to remove redundant columns.
• Compared which factors (dragons, towers, first blood, vision) were linked to winning in each year, and documented why a direct cross-year comparison was unreliable.

• Built an interactive Tableau Public dashboard comparing major stock indices from 2019 to June 2024, with every index rebased to 100.
• Designed KPI cards (best and worst performer this year, strongest growth since 2019, deepest fall), a return against volatility scatter plot, an annual returns table and an Asia-Pacific map with written analysis.

• Building a Unity rhythm game and its level editor as a solo project. Profiled the editor and cut repaint time on the densest chart from 119.8 ms to 14.9 ms, verified with 18,125 automated comparisons.

My dissertation looked at how reproducible it is to build Gaussian16 inputs for large biological assemblies, and included a scripted workflow to reduce input errors.

A technical training programme covering Python (Flask, SQLAlchemy and Pandas), SQL with MySQL, data modelling, ETL and Power BI.

Chemistry (A*), Computer Science (A), Mathematics (A), Physics (A).