From CSV Time Tracker to Full Newsroom Operations Platform

Internal operations hub for Los Santos News — AI content generation with Claude, browser-based video-to-GIF via FFmpeg.wasm, guided story building workflow, 4-tier employee lifecycle management, Discord bot with credential delivery, and shift analytics. 15 Vercel serverless endpoints, 20+ migrations, PostgreSQL RLS, and dual-environment deployment.

ReactViteSupabaseClaude APIFFmpeg.wasmTailwind CSS

Dashboard — 6 tools, one newsroom. Select a module to start working.

100%

👤

LSN Staff

Logs in with email/pass

⚡

Vite + React

SPA with hash routing

🔐

Supabase Auth

JWT, 24h TTL

🛡️

Row-Level Security

4-tier role enforcement

🗄️

PostgreSQL

Roster + shifts + profiles

📄

CSV Parser

Dual-format, encoding fallback

⏰

Shift Builder

Pair events + flag anomalies

📊

Analytics

Recharts, 7d-90d ranges

Phase 0•Core Platform•Tap a node for details

Key component

Infrastructure

Data flow

Core Platform

Phase 0

The foundation — Supabase Auth with JWT (24h TTL), PostgreSQL Row-Level Security enforcing a 4-tier role hierarchy (Admin > Editor > Creator > Viewer), employee roster with Steam ID matching, and CSV-based shift analytics with anomaly detection.

What Ships

→Supabase Auth with email/password + forced password reset gate

→Row-Level Security — 4-tier enforcement via is_admin(), is_editor_or_admin(), get_my_role()

→Roster — employee directory with rank, certs, callsigns, Steam name/hex, Discord username

→Time Tracker — CSV upload with dual-format parsing (clock events + paired in/out)

→Shift Builder — clock in/out pairing with anomaly flags (>4h long, <30m short, missing clock-out)

→Analytics — Recharts area/bar charts, 7d/14d/30d/90d/custom ranges, color-coded performance tiers

→3-pass auto name matching (exact, normalized case/spacing, reversed name order)

→localStorage fallback with Supabase cloud sync (fast offline, authoritative online)

Impact

LSN staff have secure, role-isolated access to shift data and employee records. Multi-tenancy is structurally enforced at the database level — data leaks are impossible by design.

Capability Comparison

	Phase 0: Core Platform	Phase 1: Content Engine	Phase 2: Production Bay	Phase 3: Admin & Discord	Phase 4: Voice Rewrite	Level 1: Basic RAG	Level 2: Multi-Source RAG	Level 3: Hybrid + Feedback RAG	Level 4: Agentic RAG
Tools Available	Roster + Time Tracker	+ Content Tools + Field Kit + Library	+ Story Builder + Edit Bay (Video Clipper)	+ Admin Panel + Discord Bot	+ Voice rewrite	+ pgvector RAG	+ Knowledge docs	+ Story arcs + Locations	+ AI query planning
API Endpoints	0 (client-only)	4 (generate, fieldkit, image, audit)	4 (same — new tools are client-side)	15 (+8 admin + Discord interactions)	15 (same)	16 (+embed)	17 (+chunk)	18 (+arcs)	19 (+quality gate)
AI Integration	None	Claude API (dual output)	Claude API (via Story Builder)	Claude API + Discord bot responses	Claude + voice rules	+ vector similarity	+ multi-source RAG	+ hybrid search + feedback	+ agentic retrieval
Auth & Security	Supabase Auth + RLS	Same + rate limiting	Same + audit relay	Same + Ed25519 + immutable audit log	Same	Same + embed key	Same	Same	Same
Content Types	Shift data only	Articles, posts, field reports	+ GIFs, overlays, media	+ User profiles, credentials	Same	Same	+ Knowledge chunks	+ Story arcs	+ Quality scores
Migrations	001-008	009-014	015-016	017-020	020 (same)	021	022-023	024-025	026
Discord	None	None	None	Bot + slash commands + credential threads	Same	Same	Same	Same	Same
Data Sources	—	—	—	—	0 (prompt only)	2 (Library + Roster)	4+ (+ Knowledge + Lore)	6+ (+ Locations + Arcs)	6+ (AI-selected)
Search Type	—	—	—	—	None	Vector similarity	Vector similarity	Hybrid (vector + keyword)	Agentic (AI-driven)
RAG Learning	—	—	—	—	Static	Static	Static	Feedback loop	Self-evaluation
Token Budget	—	—	—	—	+~200	+~800-1200	+~1000-1500	+~1200-1800	+~2000-3000
Latency Impact	—	—	—	—	+0ms	+250-500ms	+300-600ms	+400-800ms	+1-3s