Systems Built at work and at home

Things I’ve built and run.

Code I wrote because the job needed it, from Veritone to CloudLogics, and one app I wrote for the studio.

Case study · Built at Veritone

A production recorder for live broadcast streams built to run around the clock. It detects SRT or HLS from the URL, records in a single FFmpeg pass, cuts segments at whatever length the user sets, ships them to S3, and restarts itself when audio or timestamps go wrong.

  1. srt:// or .m3u8

    Input

    srt:// or https://*.m3u8, detected from the URL

  2. SRTHandler / HLSHandler

    Protocol

    SRTHandler or HLSHandler, one base class

  3. ffmpeg -f segment

    FFmpeg

    single pass: buffer, flags, encode or copy, segment

  4. s3://bucket/

    Output

    MP4/TS segments to S3, at a length the user sets

Monitor watches the whole run for AAC errors, timestamp drift and odd file sizes, and restarts it when something goes wrong.

  1. Command: ingest --source srt://203.0.113.24:9000 --segment-seconds 120 --bucket s3://example-ingest/ch7/
  2. 14:00:00 protocol: SRT, detected from the URL
  3. 14:00:00 srt: latency 2000 ms, caller mode
  4. 14:00:01 srt: connected to 203.0.113.24:9000
  5. 14:00:02 input: H.264 1920×1080, AAC 5.1 at 48 kHz
  6. 14:00:02 audio: 5.1 folded to stereo
  7. 14:00:02 ffmpeg: one pass, -copyts, 120 s segments on keyframes
  8. 14:02:00 segment: ch7_20250314T140000Z.ts closed, 41.8 MB
  9. 14:02:02 s3: ch7_20250314T140000Z.ts uploaded
  10. 14:04:00 segment: ch7_20250314T140200Z.ts closed, 41.6 MB
  11. 14:04:01 s3: ch7_20250314T140200Z.ts uploaded
  12. 14:04:30 monitor: AAC errors 0, drift 0 ms, sizes OK
  13. 14:06:00 segment: ch7_20250314T140400Z.ts closed, 41.9 MB
  14. 14:06:02 s3: ch7_20250314T140400Z.ts uploaded
A mock run with a made-up address and bucket. The segment length is whatever the user sets; this run uses two minutes. From S3, other AI engines pick the segments up for processing.
  • Python
  • FFmpeg
  • SRT
  • HLS
  • AWS S3
  • Kubernetes
Read the case study

Personal · open source

Synklok

A free macOS menu-bar app that keeps an RME M-1620 Pro D’s sample rate in sync with its MADI input. There’s a Windows version too, in the tray.

Tested by Harold Wendell Sanders.

  • macOS
  • Windows
  • MADI
  • Open source
Synklok’s menu: Connected, device rate 96.0 kHz, MADI input 96.0 kHz, MADI sync Locked, and three recent rate changes between 48 and 96 kHz.
The menu, connected to the pretend M-1620 I test with. Every rate change is logged.

More

Other systems.

  1. Built at CloudLogics

    AI-assisted support intake

    Claude and Claude Code workflows, with MCP connectors to the CRM and Jira, that turn customer requests into structured tickets and engineering escalations.

    Claude · Claude Code · MCP · HubSpot · Jira

  2. Built at CloudLogics

    Customer documentation site

    The docs our customers use, built on Astro Starlight and self-hosted on Coolify.

    Astro Starlight · Coolify

  3. Built at Veritone

    Support automation

    Incoming Salesforce cases ranked by sentiment, with Gemini drafting replies from tagged knowledge articles. I wrote the Salesforce–Google API integration in Apex while managing Veritone’s global support team. It helped the team reply faster.

    Salesforce Apex · Gemini API · Google APIs

  4. Built at Veritone

    Media monitoring dashboards

    Qlik dashboards on SQL and GraphQL that caught client issues in real time and fed our account health reviews.

    Qlik · SQL · GraphQL

Working tools

What I build with.

What I actually use, nothing I only read about.

Languages
  • Python
  • SQL
  • Apex
Media & streaming
  • FFmpeg
  • SRT
  • HLS
  • AWS MediaConnect
  • MediaLive
Cloud & infrastructure
  • OpenStack
  • KVM
  • Ceph
  • Kubernetes
  • Linux
  • Docker
  • AWS S3
  • S3-compatible object storage
  • IPsec VPN (strongSwan)
Platforms & APIs
  • HubSpot
  • Jira
  • Salesforce
  • Amazon Connect
  • GraphQL
  • Qlik
AI & automation
  • Claude Code
  • MCP
  • Gemini API

Contact

Compare notes.

If you work on streaming pipelines, cloud migrations or the tools around them, I’m always happy to talk shop.