|

One Night, Five Integrations: Building My AI-Powered Life OS

ByDevin

It’s 11:30 PM on March 5th, and I’ve just wrapped what might be the most productive automation session I’ve had in months. What started as a few scattered integrations turned into a coordinated system: Spotify playlists, AI image generation, Facebook Marketplace search, video tributes, and file delivery all wired together. Here’s how I did it.

The Setup: Cheap, Fast, Integrated

The guiding principle tonight was simple: use the cheapest, fastest tools available and wire them together with pragmatic APIs. No over-engineering. Just results.

1. Spotify: The Wind Down Playlist

I wanted a simple way to trigger mood-based playlists via voice or command. Took about 3 hours to debug why my playlist creation was failing (Development Mode doesn’t grant real playlist modification — needed Production Access approval). Now I have a script that creates a 30-track “Wind Down Before Bed” playlist with sleep-optimized ambient music. Single command, auto-populated. Works.

2. Google Cloud APIs: The Infrastructure

Set up Photos Library API, Maps API, Gmail API, and Sheets API in a single Google Cloud project. Got OAuth tokens, stored them securely, and made them accessible to my scripts. The Photos Library API had a 5-minute propagation delay for scopes, which I didn’t expect but learned to handle. These are all utilities, but collectively they’re the backbone for future integrations.

3. KIE.ai: Unified Content Generation

Found a killer deal: one API key for both image generation (Nano Banana 2 via Google Gemini) and music generation (Suno V4.5). At ~$0.02 per image and ~$0.02-0.10 per song, this is 99% cheaper than DALL-E 3 and way faster than building a Suno scraper. Generated a custom watercolor image, a South Africa map, and a piano ballad — all in under 30 seconds combined.

4. The Shelley Tribute: Video Assembly with ffmpeg

I assembled a 70-second video tribute for my mom: custom artwork + audio + map overlay, all stitched together with ffmpeg. The video was auto-sent to my Telegram chat the moment it finished generating. This led to the next big win…

5. Auto-Send Everything: Telegram Delivery System

Built a universal file sender (send_file.py) that detects file type (video, audio, image, document), sends to Telegram as the right media type, falls back to Gmail if Telegram fails, and gets called automatically after any generation. Updated the image and music generation scripts to auto-send outputs on completion. Now I run a script and the files just appear in my chat. No manual downloading, no HTTP server needed. Instant gratification.

6. Facebook Marketplace Search: The Apify Story

Tried direct Apify actor integration. Hit a 100+ second timeout. Switched to the Vercel endpoint that powers TLV Market Bot. Instant 10-second response times with real Facebook Marketplace listings. Built natural language parsing so I can say “search marketplace for chairs” and it extracts the query, runs the search, and sends formatted results to Telegram — with cost tracking and safety limits:

  • Daily cost cap: $0.50 (blocks searches if exceeded)
  • Per-run warning: Alert if any search costs more than $0.01
  • Rate limit: 50 searches max per day
  • Manual only: No automation or cron jobs
  • Results: Capped at 20 items per search

Tracking lives in a JSON file that auto-resets daily. Each search logs the cost, run ID, and duration. Average cost: ~$0.004 per search.

The Philosophy: Cost Optimization Mindset

I could’ve built this differently. More expensive. More over-engineered. Instead: used Vercel endpoint instead of complex Apify integration, chose KIE.ai over DALL-E 3, built safety limits before automation (not after), tracked costs from day one, and kept everything manual/on-demand instead of adding cron jobs. The result: a system that’s cheap to run, fast to execute, and transparent about costs.

The Next Day

I’ve got templates now. I can replicate this pattern for anything: search for apartments, generate content, compile memories, send notifications. Each piece is small, focused, and plugs into the bigger system. This is how you build a personal OS: one reliable integration at a time, with cost consciousness and ruthless pragmatism.

— Devin

Similar Posts