In practice

Selected work.

Projects from the founder's previous enterprise roles, anonymized, plus one Anchor Data Labs prototype. Each case gives the role, the contribution, and the outcome.

5d→1h Intent-doc + SQL prep, team-reported

Delivered

Snowflake
Python
Claude API

AI Knowledge Engineering · Financial ServicesUK

A SQL-support agent's knowledge base, re-engineered.

Situation: A team at a major financial institution built an agent to turn analyst requests into SQL, but its knowledge design re-read 1,000+ SQL files for every prompt and answered inconsistently.
Role: Re-engineered the knowledge base around structured Snowflake retrieval, replacing 1,000+ file reads with two to four governed queries. Contributed to the Python bridge exposing approved read functions, and tested the request-to-SQL workflow.
Outcome: The project lead reported the team's automation cut intent-document and SQL preparation from about five days of human dependency to roughly an hour. The retrieval redesign was my contribution: repeated large file reads replaced by targeted database queries.

24h→min Internal data-governance guidance turnaround

Delivered

Claude Skill
GitHub

AI Knowledge Engineering · Financial ServicesUK

Data governance guidance inside Claude projects.

Situation: Product, process, analyst and engineering teams needed governance guidance, on naming, modelling, and how to move, store and deliver data to standard, and each question typically waited about 24 hours on a human consultant.
Role: Started and led the knowledge hub for an internal data-guidance agent, built from the Data Management team's consulting work: naming conventions, data models, pipeline and infrastructure choices, exceptions and anti-patterns, and which team to route to. Kept it versioned in GitHub with change management, and packaged it as a Claude Skill teams could use while coding.
Outcome: Teams got governance guidance in minutes instead of a roughly 24-hour wait, with routes back to the owning teams for anything deeper.

0 Reporting outages, data-centre exit migrations

Delivered

Migration · Financial ServicesUK

Data-centre exits, on-prem to cloud. Reporting never went dark.

Situation: A major financial institution was exiting data centres, moving databases from on-prem to cloud for application processing, complaints and document storage, and in places from Oracle to Salesforce and Snowflake. Every report that sourced from them had to be re-mapped, rebuilt, or pointed at the new source, and some had to be built fresh.
Role: Led the analytics side of the report migrations: impact assessment, dataset redesign, re-mapping and cutover sequencing. Led delivery across teams during peak migration periods, and owned some data-centre exit migrations end to end, coordinating report owners, data engineers and analysts across teams.
Outcome: Moved reports at volume across the exits, each re-mapped or rebuilt and validated before cutover, so the business kept its reporting throughout. Smaller migrations of 10–15 reports each ran steadily alongside.

2,000 Repositories migrated to the cloud

Delivered

Migration · Source Control & DevOpsUK

Moved a 2,000-repo GitHub estate to the cloud.

Situation: A data department's entire GitHub estate, 2,000 repositories, ran on the company's own GitHub servers and had to move to GitHub's cloud.
Role: Owned the migration end to end: built a custom script to back up, standardize, and move every repository, set the standards for the new environment, and ran onboarding for new hires, coordinated with a platform team we did not sit on.
Outcome: All 2,000 repositories moved to the cloud, with teams productive throughout.

50h Manual prep removed, monthly

Delivered

Google Apps Script
Python

Automation · Financial ServicesUSA

Automated spreadsheet preparation and risk-assessment QA.

Situation: Banking teams were spending about 50 hours a month on manual spreadsheet prep, with recurring manual QA on risk assessments on top.
Role: Built a Google Apps Script automation for the spreadsheet prep and a Python automation for risk-assessment QA.
Outcome: About 50 hours a month of manual prep removed, plus roughly 20 hours a quarter saved on risk-assessment QA.

~100h Manual financial-planning effort saved, quarterly

Delivered

Python
Power BI
SQL Server

Machine Learning · MiningUSA

Custom ML for operations, and a reporting rebuild that cut planning time.

Situation: A mining company needed operational analysis and relied on spreadsheet-based financial planning.
Role: Developed and evaluated a machine-learning model for operational time-series analysis, and replaced the spreadsheet-based financial planning with Power BI and SQL Server.
Outcome: The reporting rebuild saved roughly 100 hours a quarter.

Prototype WhatsApp shop-floor tracker, built and sandbox-tested

Prototype · pilot on hold

Next.js
PostgreSQL
Meta WhatsApp Cloud API

Automation · Manufacturing (SMB)India

A factory floor that reports itself over WhatsApp.

Situation: A Mumbai manufacturer's managers were chasing shop-floor workers by hand for the status of each job.
Role: Designed and built a work-in-progress tracker: workers report each batch handoff from any phone through an automated WhatsApp menu, no typing and no app to install, and the system records the update, adjusts stock, and triggers the next message, while owners watch live stock, scrap, and stuck orders on a secure dashboard.
Outcome: Built and tested end to end on real phones in a sandbox: every message is signature-checked, a double tap cannot record a handoff twice, and the handoff history cannot be edited. The pilot is on hold pending the client; production validation is part of the next stage.

These are the kinds of problems our service lines solve. If one looks like yours, tell us about your project.