Steam trading-bot service
A separate service that sends trade offers to customers with steam-tradeoffer-manager and follows each one through to an accepted trade.
Software developer · Denmark
PeterJespersen
I build products end to end — storefronts, trading bots, data platforms, music apps. Below: every commit I've made, on two GitHub accounts, as a city.
2017 → today
contributions · 0 active days · 2 accounts · 1 developer
github.com/piijt + github.com/peterhjespersen
Two accounts.
One graph.
Side projects on my personal account, product work on my work account. Merged day by day, every year since 2017.
live Case study · 2019 → today · solo-built
Trade your CS2, DOTA2, RUST and TF2 skins for World of Warcraft gold. I designed, built and run the whole platform — storefront, API, Steam trading bots and pricing engine.
- live in production since
- 2019
- orders completed
- 7,000+
- gold given away in giveaways
- 25M
- Trustpilot reviews
- 158
The core loop
One trade, four steps, zero middlemen.
Under the hood
Every layer, built and run by me.
- StorefrontNuxt 4 · Vue 3 · TS
- APIExpress · MongoDB · pricing
- Trade botssteam-tradeoffer-manager
- SteamSteam Web API
Pricing engine
Converts Steam market prices into gold per game version and region, layered with event bonus percentages and a +10% bucket bonus for selecting more than 10 items.
Stacked items
Rust-style stacked items are handled as quantities, so a stack is priced, selected and traded correctly instead of as a single skin.
- Nuxt 4
- Vue 3
- TypeScript
- Node.js
- Express
- MongoDB
- Steam Web API
- Docker
Side project · music
flow
Your music, saved and always playing.
A self-hosted music library: search YouTube or import a Spotify playlist, build playlists, and play it all from your phone with the screen off. That last part drives the entire architecture.
- Nuxt 4 SPA
- shadcn-vue
- Tailwind 4
- FastAPI
- yt-dlp
- MongoDB
- MediaSession API
- Docker + Caddy
- 0
- YouTube iframesa locked phone kills them
- 1
- <audio> elementmounted once, never re-created
- 206
- Partial ContentRange support written by hand
- 0
- Node runtimes in prodstatic SPA served by FastAPI
9:41
Good evening
Recently addedShow all
Neon TidesHarbor Lights
Friday 9 October
9:41
Neon TidesHarbor Lights
0:37−3:05
Live recreation · tap anything
Background playback is the whole point.
A tiny synth built live with WebAudio, kept quiet. While it plays, the visualizer reads a real AnalyserNode instead of the synthetic beat.
Architecture
From a search box to a locked phone.
Every box is a decision the lock screen forced. Tap one for the detail that makes it work.
YouTube
Search that pages properly
yt-dlp has no continuation cursor, so a deeper page means asking for a larger N. Each query is cached for 10 minutes, over-read by 40, and merged rather than replaced, so pages never overlap when YouTube reshuffles.
ytsearchN: · LOOKAHEAD 40 · merge, don't replace- audio bytes
- matching · control
Spotify import
Matching songs, not guessing.
Each Spotify track is searched on YouTube and every candidate is scored, with duration weighted most because a different length almost always means a different recording. Below 0.40 is reported as not found; below 0.72 is flagged for you to check. This runs the real formula.
Spotify playlistLate Night Drives5 songs · syncs in place on re-import
- 1Neon TidesHarbor Lights · 3:42scoring
- 2Paper SatellitesMira Vale · 4:05queued
- 3Slow BloomOdd Fjord · 3:18queued
- 4GlasshouseKite Theory · 2:56queued
- 5Untitled 7Sora Wren · 3:33queued
3 matched · 1 to check · 1 not found → a matching flow playlist, name and cover included
YouTube candidates forNeon Tides · Harbor Lights · 3:42
- Harbor Lights - Neon Tides (Official Audio)Harbor Lights · 3:43 Δ1s–
- Neon Tides (Live at the Harbour)Harbor Lights · 4:28 Δ46s–−.18 live
- Neon Tides 1 hour loopchill loops · 60:00 Δ3378s–−.30 compilation
duration ·45title ·35artist ·20| 0.40 · 0.72
Play before you save
Every search result plays as a preview within seconds. Add promotes it in place, no second download; anything you don't keep is swept after a few hours.
Video, slaved to the audio
Opt-in video per track. The <audio> stays the clock: big gaps snap into place, small drift is absorbed by nudging playbackRate a few percent. Toggling never interrupts playback.
Built for other people
Follow people, share your library per follower, follow their playlists, see friend activity and join them at their position. Messaging needs a follow both ways.
Real accounts
argon2 hashes in MongoDB. The admin is seeded on startup; everyone else joins through an expiring invite link generated on the Admin page.
One container, one origin
Docker Compose for the app and MongoDB, Caddy for automatic HTTPS. The SPA is prebuilt and served by FastAPI, so there is no Node runtime in production.
Installs like an app
Add to home screen over HTTPS and it behaves like a native player. The service worker caches only the shell and hashed assets, never /api/.
Product · data · SaaS
A multi-tenant data workspace I designed and built end to end. Point it at MySQL, Postgres or MongoDB, save a query, and every run is captured as an immutable, versioned snapshot: diffable, schedulable and ready to chart. Every row, query and API key is fenced off by organization.
- Nuxt 3
- Vue 3
- TypeScript
- Tailwind
- shadcn-vue
- ECharts
- Express
- Mongoose
- MongoDB 7
- MySQL 8
- Postgres 16
- Stripe
- Docker
- Caddy
- 013 enginesMySQL, Postgres and MongoDB connections, plus HTTP APIs and CSV / JSON uploads.
- 02VersionedSchema inferred, rows coerced, deduplicated by dataHash. Every run is a new snapshot.
- 03ScheduledConnection-backed snapshots auto-refresh on an interval, hands-free.
- 04Multi-tenantorgId on every model and in the JWT. Four roles, invites, Stripe trials.
SELECT region, COUNT(*) AS orders, SUM(total) AS revenue, status FROM orders GROUP BY region, status; read-only ✓0/32 rows
Click a source to push a burst of rows · click a card to open that version
localhost:3000/dashboards/ops-overviewlive
v12
v11→v12+0−0~112:00 · #bf36ce1
| region | orders | revenue | status | |
|---|---|---|---|---|
| EU-West | 289 | 10.7k | paid | |
| APAC | 277 | 12.3k | paid | |
| Nordics | 133 | 5.4k | paid | |
| ~ | US-West | refunded | ||
| LATAM | 208 | 9.2k | paid | |
| MEA | 84 | 3.2k | paid |
schema region:string · orders:int · revenue:float · status:enum
owneradmineditorviewer
// decoded JWT { "sub": "usr_4c1e", "role": "editor", "orgId": "org_acme" }
Click an organization to sign in as it
What's next
Let's build
something loud.
© 2026 Peter Jespersen