Inspiration
I work with IT environments where a single support case can mean opening ScreenConnect, Microsoft Intune, Entra, Atera, an internal device inventory, and sometimes a separate company tenant. The technical data exists, but the operator has to reconstruct the situation manually every time. I wanted the first screen to answer a more human question: who is sitting where, which computer are they using, is it online, and does it need attention?
What it does
OneOps is a local-first IT operations control plane for multi-company teams. Its Live Workplace turns the office floor plan into an operational interface. Each desk can show its employee, computer, live presence, session duration, compliance state, and risk.
Selecting a desk opens one reconciled record while preserving the source of every fact. A service desk ticket is linked to the affected person and endpoint, company switching updates the entire workspace, integration health stays visible, and uncertain cross-provider matches go to a human review queue instead of being merged blindly.
The Hackathon build includes:
- An interactive desktop and mobile workplace map
- Multi-company scope switching
- Unified device records and global search
- A prioritised service desk queue
- Connector health, stale-data states, and manual refresh
- Microsoft Graph presence and ScreenConnect bridge adapters
- Deterministic reconciliation with source provenance
- A review workflow for ambiguous identity matches
How I built it
The frontend is React and TypeScript with a deliberately dense operational design rather than a marketing dashboard. An Express API collects provider data in parallel behind a provider-neutral connector contract. Shared in-flight requests, TTL caching, stale-on-error fallback, sanitised diagnostics, and explicit live/demo modes keep provider failures understandable.
The default dataset is synthetic so judges can run every workflow without company credentials. Live connectors are server-side and read-only by design; secrets are never exposed through Vite or browser storage.
I used Codex with GPT-5.6 throughout the build. Codex inspected the evolving repository, implemented the React/TypeScript interface and Express connector layer, wrote reconciliation and cache tests, and helped conduct six task-based user-testing rounds. GPT-5.6 was most valuable when reasoning across provider identity, stale data, least-privilege boundaries, and mobile workflows. I kept product decisions human-led and verified changes through the running app, automated tests, production builds, screenshots, and release checks.
Challenges
The hardest problem was not drawing a floor plan. It was deciding when records from unrelated systems represent the same physical device without hiding uncertainty. OneOps separates canonical identity from provider evidence, retains provenance, and sends low-confidence candidates to review.
The second challenge was honest failure handling. A unified dashboard becomes dangerous if stale Microsoft data looks live, so the API exposes connector timestamps, health, configuration errors, and fallback state directly to the UI.
Accomplishments
- A complete spatial-to-incident journey from desk A-04 to an urgent BitLocker ticket
- Seventeen automated tests for API contracts, caching, fallback, mapping, telemetry sanitisation, and reconciliation
- Six user-testing rounds across first-time judges, IT operators, workplace leads, and mobile service desk use
- Responsive verification at a real 390 by 844 mobile viewport
- A reproducible Docker production build and public v1.1.1 release
- Sensitive write actions remain intentionally disabled until authentication and immutable auditing are present
What I learned
A useful unified operations product should not erase the systems behind it. Operators need one view, but they also need source traceability, freshness, and a clear boundary between evidence and inference. The floor plan became valuable only after those trust signals were designed into it.
What's next
The next milestone is a non-production pilot with Microsoft and ScreenConnect, validating twenty known desk-device-user matches before enabling read-only Atera data. After that I would add role-based access, step-up authentication, immutable audit events, and carefully gated remote-support launches.
Built With
- codex
- docker
- express.js
- gpt-5.6
- microsoft-graph
- react
- screenconnect
- typescript
- vite
- vitest