Sprint showcase, anonymized

Seven sprints, one client

This is the work I did for a US investment analytics platform between August and October 2026. Analysts upload Python scripts, the platform runs them each night against market data, and members read return cards, fund drill downs and daily emails. I removed every name, domain, repository and billing detail. The counts come from the delivered reports, the written updates and the merged pull requests. Every sprint below was scoped, reported and handed over in writing.

Length: 2 to 9 days per sprint, 7 sprints
Scope: Features, data fixes, infrastructure and security findings
Stack: Python and Django, React and TypeScript, Postgres, AWS containers, Terraform, GitHub Actions
Outcome: 16 pull requests, 336 commits, tests from 52 to 922

Five clear glass blocks in a row on a black steel bar. The orange channel inside each block runs a little further than in the one before, from barely started to complete.

[2]Timeline

Sprint 18, nine days

Before day 1Request
The written scope
The client sent a list of requests in a chat message. Within one working day I replied with each item in their words, what I would deliver, how I would check it and the hours per part.
Artifacts. Scope
written
Day 1Kickoff
Kickoff note and scope report
A short kickoff message with the hours per part and the delivery day. The same evening the full scope followed as a report, with two findings that needed a decision first.
Artifacts. Kickoff note
Scope report
Day 2Build
All eight builds tested and live
Every item on its own branch with tests, reviewed and deployed. The drill down now reads about 35 times fewer price rows per page. After the deploy one query made a fund page slow, and I cut its first load from 8 to 12 seconds to 1.7 to 4.7.
Artifacts. 6 pull requests
3 migrations
Days 3 to 8Watch
Nightly checks on the live site
A watcher I left running proved that the nightly pipeline ran out of memory on 14 of 30 logged nights. I also found API keys in the code of 72 stored scripts and limited script code to staff the same day. A mid week note reported both.
Artifacts. Progress note
Watcher log
Day 9Deliver
Report, evidence and handover
A delivery message, a report with 13 sections and 22 evidence references, a compliance sheet and a time statement. Before sending, a validation pass checked 388 statements against raw data. After the deploy a second pass checked 441.
Artifacts. Report
Compliance sheet
Time statement

[3]Sample artifacts

Written samples

Two short excerpts from real messages, anonymized.

A finding sent before the deploy, sprint 17BAnonymized

One item cannot wait for the rest of the sprint. An old test script prints the platform's own settings into its error text, with real values. The platform shows that text to any signed in user, and anyone can register.

The code side is fixed on the branch. Error text is now shown to staff only, and any secret value is replaced before an error is stored or logged.

One rotation signs every user out once, so it needs your go ahead and a quiet half hour. I have changed nothing on production.

Michael

The close of a report, sprint 13Anonymized

Everything above is built, committed and proven against real data and a running application. None of it is deployed to your production environment yet, and I want that stated plainly rather than buried.

Michael

[4]Outcome

What shipped

Three month return on cards, tables, CSV exports, the daily email and its PDF, checked to four decimals
Sticky headers on three long tables, tested at desktop and phone width
A drill down of 1,800 rows, paged, sorted and searched on the server
Staff reruns of a script for an earlier date, limited to the script sandbox
A database disk moved from 10 GB to 20 GB on faster storage, with the database online
A fund page first load cut from 8 to 12 seconds to 1.7 to 4.7 seconds
Script source code limited to staff after API keys turned up in 72 stored scripts

[5]At a glance

The last five sprints, at a glance

The same client, most recent first. Across these five the backend test suite grew from 52 tests to 922.

Every sprint

A finding that changes the client's risk goes first, before the feature list.
I fix the cause, not the symptom, and I say what the cause was in plain words.
Every figure in a report carries a reference to its evidence. I mark estimates as estimates and state my own corrections.
Nothing touches production without a snapshot first. Anything that changes who sees what, or rotates a secret, waits for written approval.
Reviews, merges and validation the client did not ask for are not on the time statement.

Seven sprints, one client

Sprint 18, nine days

The written scope

Kickoff note and scope report

All eight builds tested and live

Nightly checks on the live site

Report, evidence and handover

Written samples

What shipped

The last five sprints, at a glance

If an agent or LLM feature breaks in production, send me a written brief.