Bournemouth, UK · Data engineer

Annas Mazhar

I build AWS data lakes — Glue, Athena, S3 — and, outside work hours, tools that prove their claims. Every repo ships its tests, its evidence file, and the failure modes it does not hide.

The short version

17merged PRs
in other people's repos
8of those merged
in repos over 6k stars
5open-source tools
published and tested
0factors promoted
on real data, honestly

Tools

factorproof

Factor research with the controls quant work usually skips: purged walk-forward CV, Newey-West HAC t-stats, Benjamini-Hochberg FDR, and a promotion gate that refuses noise.

14 coins · 21,332 daily bars all 14 factors rejected 201 tests

replayproof

Offline regression testing for LLM agents. Record a run once, replay it forever with zero API keys, assert tool-call contracts, and diff behaviour when you swap the model.

zero API keys token-cost gate CI on 3.11 and 3.12

fitsproof

A calibrated cost model and an enforced memory budget in front of a from-scratch NumPy inference engine with KV cache and quantisation. When the model does not fit, it says so loudly.

no silent OOM NumPy engine

pytest-duckdb

Pytest plugin for SQL pipeline testing with DuckDB: load fixtures, run queries, snapshot results. Listed in the official pytest plugin index and used in third-party suites.

PyPI official plugin list Python 3.9 to 3.13

pyspark-mcp

An MCP server for data engineers: SQL-to-PySpark transpilation, AWS Glue job generation, and Spark optimisation, usable from any MCP client.

MCP registry Glue · Spark

Merged upstream