Spiros Raptis Independent systems engineer GitHub

Independent systems engineer / Thessaloniki

Spiros Raptis

Products and systems that hold up in operation.

I build and operate product apps, local-first utilities, agentic workflows, and self-hosted infrastructure. The work continues past launch: careful migrations, recovery paths, and verified outcomes.

See the product work Public source

A dedicated domain is a small detail with a large implication: the project now owns its production identity.

Current release
v0.18.0
Primary domain
kalathitimon.com
Continuity
Compatibility mirror retained

Selected work / case notes

Products with an operating story, not only a launch story.

The work spans consumer products, native operations, and private edge systems. Each case includes the migration, failure condition, and continuity work that keeps it useful after release day.

Case 01 / Independent product

React / same-origin services / dual-target publishing
Production / v0.18.0
kalathitimon.com

Independent product ownership

A useful tool graduated into its own production surface.

The domain move changed more than the address.

Καλάθι Τιμών now serves the app, PHP APIs, images, daily opportunities, and compact basket links from kalathitimon.com. The hourly pipeline publishes the same validated catalogue atomically to the new primary domain and the established compatibility mirror.

Production domain
kalathitimon.com
Current release
v0.18.0
Sharing path
Same-origin /s/ links
Continuity path
Compatibility mirror
Open kalathitimon.com Review source
Migration constraint
Move the product without breaking existing baskets or splitting catalogue state.
Engineering response
Use config-driven paths and publish one validated snapshot to both destinations.
Operational outcome
The product has an independent identity while established links keep working.

Case 02 / Native operations

Fleetlight

A menu-bar control surface for a mixed Mac and Linux fleet.

Reachability, service probes, guarded maintenance, and local history distinguish a command that was sent from an outcome that was verified.

Risk
Remote state can be stale or ambiguous.
Response
Probe before action; verify after it.
Review repository

Case 03 / Private edge system

Camera Sentinel

Private monitoring that treats feed health as first-class data.

Local video, motion zones, event history, bounded alerts, and guarded recovery keep a failed or frozen feed from looking like a quiet scene.

Risk
A healthy-looking interface can hide a dead feed.
Response
Monitor the signal, not only motion events.
Review repository

Recent operating record

Work that moved from idea to maintained behavior.

Public-safe summaries only. Private hostnames, workflow data, personal metrics, and infrastructure details stay out.

01 Product

Καλάθι Τιμών moved to its own domain.

Version 0.18.0 gave the product an independent production identity while preserving the original compatibility mirror.

Open

02 Operations

Fleet actions gained outcome receipts.

Guarded actions and post-action probes now separate intent from confirmed machine state.

Repo

03 Automation

Review workflows made the last step explicit.

Evidence capture, recovery markers, and guarded completion keep automation inspectable at the risky edge.

Repo

04 Media

Plex gained a quieter front door.

A self-hosted web and Android surface keeps remote playback and subtitle-heavy use central.

Repo

Engineering practice

Dependability is an operating model.

A reliable system is not one that never fails. It is one that exposes the failure, contains the damage, preserves a recovery path, and makes the final state easy to verify.

  1. 01

    Diagnose the actual state.

    Start with probes, logs, page state, and the user-visible symptom before changing configuration.

    Observe / reconcile / decide
  2. 02

    Bound the risky action.

    Keep permissions narrow and irreversible transitions explicit enough for a person to inspect.

    Guard / confirm / execute
  3. 03

    Make recovery routine.

    Snapshots, receipts, local history, and bounded retries turn failure into an ordinary operating state.

    Record / retry / restore
  4. 04

    Verify at the surface.

    A successful command is only an intermediate event. The product or machine state is the result.

    Probe / compare / close

The professional standard

Leave the system easier to understand, safer to operate, and less dependent on memory than it was before.

Open systems index

A broader record of shipped work.

Selected public repositories across native apps, local-first tools, private media, agentic automation, and earlier research tooling.

01 / Learning app

Italian Coach

Adaptive, local-first Italian practice with mistake review and optional self-hosted sync.

macOS / Android

Repository
PX

02 / Self-hosted media

Plex Open

Browser and Android playback built around subtitles, remote access, and living-room use.

Web / Android

Repository
AU

03 / Agentic operations

Review Automation

Evidence-led workflows with model controls, state recovery, guarded completion, and companion apps.

Agents / macOS / iOS

Repository
SC

04 / Native utility

Screenshot Bridge

A Swift menu-bar utility that moves screenshots directly into a Codex workflow.

Swift / macOS

Repository
RS

05 / Earlier work

Past Markets Lab

Earlier local research and execution tooling; no longer an active trading focus.

Rust / Local dashboards

Repository

Working stack

Start with the real operating condition

A dependable system begins with an honest account of what can fail.

Bring the repository, the constraint, and the behavior that needs to become observable, recoverable, or easier to operate.

Open GitHub Message @srdevb