A US hotel-intelligence company
Market data collected every night for a hotel-intelligence company
Collectors with proxy rotation feeding MariaDB, running every night since 2025. 8M+ requests a month against the sources the collectors watch.
An independent trader. March 2024 to May 2026.
Market-data tooling for a trader, on a two-year retainer
A trader had alerts in one format, market history in another and no way to ask what a price was when an alert fired. Five Python modules enrich the alerts from Polygon, normalise the timestamps and prepare his backtest datasets, on a retainer from March 2024 to May 2026.
Financial markets. Data.
Market-data backtesting in figures
- 26 monthson retainer, March 2024 to May 2026
- 5Python modules, from alert enrichment to backtest datasets
- Trade alerts
- Polygon market history
- Timestamp normalisation
- Price at a given time
- Backtest datasets
Screens from this system are being added.
The situation
An independent trader had alerts in one format, market history in another and no dependable way to ask what a price was at the moment an alert fired. Every backtest started with the same data wrangling, done by hand.
What we built
- A module that enriches trade alerts with historical bars from the Polygon API
- Timestamp normalisation across the alert feed and the market data, so the two line up to the minute
- Price lookup at a given time, including the edge cases around market open and close
- Dataset preparation for analysis, written so each backtest starts from the same clean table
- Walkthroughs of the code with the trader as the tooling grew, over two years of retained work
What changed
The trader spends his time on the strategy, not on the data underneath it, and kept the retainer running for 26 months.
Read next
Our own engineering
A retail catalogue kept in step with Shopify, price by price
A scheduled Python pipeline that reads Costco product pages, stores them in MongoDB and creates and updates the matching Shopify products, prices and inventory by SKU.
A real-estate researcher working Montana property records
Every lot in the Yellowstone Club, as one dataset
Every house and lot in the Yellowstone Club community, extracted from Montana's cadastral API and its HTML pages into a single CSV, with a notebook that measured how the extraction behaved. 3 days from the approved sample to the full dataset, November 2023.
Elsewhere on the site
Related
- PracticeThe Data practiceWhat the practice covers, what it costs to start, and the work behind it.
- The workDemosThe demos you can open now: the two that run in your browser, the hosted prototypes, a walkthrough on video and a live cafe site.
- The workThe five practicesBuild, Automate, Data, Growth and Talent: what each one covers, the work behind it, and what it costs.
Where to next
Have a problem in this shape?
Say so in a few sentences. The reply comes from the person who built this one.