More I have built 2025-2026

Data and research tools

Four data tools: no lost alerts, no duplicate rows, and tests that cannot cheat.

Alerts, data imports and fair tests, built alone. A home buyer stops checking sites daily, and lucky-looking ideas are caught before real money.

01 The problem

Data that changes all the time is slow to check by hand. A test that quietly uses future data, or simply gets lucky, gives false confidence.

A home buyer gets every new match and price change in one chat thread per ad, without daily site checks.
Two market strategies that only looked good were stopped before any real money was used.
A five-year test now runs in 0.08 seconds instead of about 21, so ideas can be checked again and again.

01
Alerts that come on time
A watcher checks public home listings on a schedule and sends a chat alert for each new match, price change, edit or removal.
02
Context, not raw rows
Each home alert shows the price per square meter against the area median, the ad's rank in its area, and its price history.
03
Imports without duplicates
Large public data sets are cut into time slices that load side by side. Running an import twice never adds duplicate rows.
04
Only data known then
Research tools use only data that was known at each moment, so past results are never quietly helped by the future.
05
Runs on its own
The watcher restarts by itself, tracks its own health, and sends a warning after three failed runs in a row.

Risk 01
Results that were only luck
A strategy can look good by chance. Each one is judged on a later period it was never tuned on, then re-run 10,000 times on random mixes of its own trades.
Risk 02
No lost alerts
A crash between storing data and sending could lose an alert. Each alert is stored in the same step as its data change, then sent with up to 5 retries.
Risk 03
No false 'gone' alerts
A search that stops early could make live ads look deleted. An ad is marked gone only when the search really reached its last page.

Example 01
Home listings watcher
Watches a classifieds site for homes that fit a buyer's filters, and spots old ads posted again under a new link.
Example 02
Forecast accuracy study
Scored almost five years of past weather forecasts against real airport readings, and delivered the error tables in Excel.
Example 03
Market idea test bench
Tests trading ideas on years of past prices with fees included, and shows each trade on a chart with the reason it was taken.
Example 04
Market research data import
Loads public market trade records into a database for research, with automatic retries and within the data source's request limits.

Hiring an engineering lead, or building a platform or AI agents? Tell me about it.