Skip to content
Back to Project Archive

Glassbox

Cyberpunk internal feedback terminal — employee signals, Kanban neural grid, MoM metrics, Gemini sentiment. Not VM2, not a code sandbox.

Glassbox Logo
// PROJECT_ID
glassbox
// AUTHOR
DHAATRIK CHOWDHURY
// SUBJECT
CASE STUDY
// CLASSIFICATION
PUBLIC // OPEN
// TELEMETRY
STATUS: SHIPPED // AI: BYOK_GEMINI // PRODUCT: FEEDBACK_TERMINAL
// IMPACT.LOG
// FUCKUP_LOG
RECORDED
[ FUCKUP LOG ]  I described Glassbox as a VM2 sandbox on this site while the repo shipped a cyberpunk feedback terminal — and Gemini once labeled sarcastic 'mandatory fun' feedback as POSITIVE.

SYS.STATUS: v4.2.1 shipped — feedback signals, neural grid, metrics, AI insights

Glassbox is an internal organizational feedback terminal wrapped in a cyberpunk aesthetic. Employees submit “signals,” managers organize them on a drag-and-drop neural grid, and leadership views month-over-month metrics plus optional Gemini-generated insight reports.

This page replaces stale portfolio copy that wrongly described a VM2-isolated Node.js code sandbox. Glassbox is not a security product.

What it is (scope)

SurfaceWhat you do there
New SignalSubmit department feedback; AI classifies POSITIVE/NEGATIVE/NEUTRAL at intake
Neural GridKanban board with drag-and-drop; filter by department, status, keyword
MetricsMoM volume trends (Recharts); trigger AI insight synthesis
Terminal UICyberpunk mission-control aesthetic with Motion animations

Stack: React 19, TypeScript 5.8, Vite 6, Tailwind CSS 4, Recharts, @hello-pangea/dnd, @google/genai, Express, better-sqlite3, Vitest.

Requires VITE_GEMINI_API_KEY — BYOK inference costs apply.

Who I built it for

  • Teams prototyping internal sentiment workflows before enterprise HR tooling
  • Developers exploring Gemini classification on short employee text
  • Anyone wanting a distinctive Kanban + metrics demo to fork (MIT licensed)

Not for: running untrusted JavaScript in isolation, compliance-grade HRIS without your own hardening, or cultures where sarcasm-heavy feedback needs human-only review.

Fuckups & learnings

  • Wrong product identity on my portfolio. VM2, sandbox escapes, JSON fs/network audit logs — stale fiction from an abandoned direction.
  • Sentiment misclassification on sarcasm. “Oh great, another mandatory fun Friday” scored POSITIVE once. AI labels are triage hints, not HR verdicts.
  • Kanban + live filters broke drag state. Mid-drag filter changes dropped card metadata until state keys were tightened.
  • Cyberpunk chrome adds maintenance cost. Motion layers and terminal UI increase bundle and test surface vs boring admin panels.

Honest limitations

LimitationReality
Not a security sandboxNo VM2, no untrusted script execution, no fs/network audit
BYOK GeminiYou supply API key and pay inference costs
Demo/internal shapeNot multi-tenant SaaS or production HRIS out of the box
Sentiment trust limitsSarcasm, coded language, and cultural context can fool classifiers

Deep-dive transmissions

Read in order for the full story:

  1. Why Glassbox exists — org feedback terminal, not a code sandbox — origin, scope, VM2 portfolio drift confession

Run it locally

git clone https://github.com/dhaatrik/glassbox.git
cd glassbox
npm install
# .env: VITE_GEMINI_API_KEY=your_key
npm run dev
# Opens http://localhost:3000

Tests: npm test — Vitest + React Testing Library.

Closing transmission

Submit signals. Drag the grid. Read the metrics. Glassbox turns qualitative employee input into structured, chartable data — with AI assist if you configure a key. Start with why-and-what if you want the full identity correction story.