Freelancing / job search · Own product
A personal tool on the official Upwork API: a worker copies every new posting into Postgres, my own filters run over that archive, and the matches arrive in Telegram.
The challenge
On Upwork a good posting collects proposals fast. A freelancer who opens it hours later competes with everyone who saw it first, and I can't spend the whole day refreshing the feed.
Upwork's saved searches are too coarse for how I choose work. I want to skip some countries and categories, set a minimum rate, check the client's spend, rating and payment status, and decide what happens to the many hourly postings that name no rate at all.
I also wanted history. A filter I write next month should run over everything posted this month, so the tool has to keep every posting, including the ones no filter wanted at the time.
The approach
On every sweep the worker asks the official Upwork GraphQL API for the newest postings with no search criteria and stores each one in Postgres, full API payload included. Results come newest first, so a sweep stops at the first page with nothing new, and a caught-up sweep costs a single request. Filters are plain functions over the stored rows and never call the API.
The worker stores a posting once, keyed by its Upwork id, and announces it at most once per filter. A match Telegram did not accept stays marked as undelivered, so after an outage or a bad token the pending alerts go out with the first sweep that gets through.
Some fields the schema declares required can arrive empty, and under GraphQL rules one such value drops the whole posting from the results. The query reads them from nullable fields, partial responses go to the log while the run carries on, and each row keeps the raw payload, so a renamed field leaves one column empty while the data stays in the payload.
About a third of hourly postings name no rate. A filter with a rate bound skips them unless I tick a box to keep them, and it compares against the top of the client's range: a $20–60 posting clears a $50 minimum, because $50 is inside what the client offered.
After I change a filter, Re-scan archive runs it over every stored posting on my server and marks the new matches as already announced. The feed and the match counts update straight away.
The first version dates back to 2018. By 2022 it was a Yii2 app with a polling daemon, a cleaner that kept a week of postings plus starred ones, exclusion filters, one-click search presets and manual 1–10 scores, and its database held 574,606 postings. Gotcha keeps the filters and presets that proved useful and rebuilds the rest on the official API, Postgres and Telegram.
In motion
The feed lists every posting in the archive, newest first. Each row puts the client's spend, rating and payment status next to the rate, and green badges show which of my filters picked it up.
Set keywords and categories, press Re-scan archive, and the filter runs over every stored posting on the server. The match count appears straight away, computed from the archive.
Stats rank skills by the median rate and budget across the archive, and a skill needs enough postings to be ranked at all, so a couple of outliers can't top the list.
See it in action
The outcome
Send a short brief about your project. Within 24 hours you get an estimate or a clear next step, even if we don't work together.