Blind spots cost time

Bottlenecks go unnoticed for 3–5 days. Overloaded team members stay invisible until the townhall.

  • 🔴 Red projects detected late → firefighting mode
  • 📉 Delayed dependencies cascade across portfolios
−32% productivity leak
🗣️

Meeting overload

Teams spend 10+ hours/week just asking “what’s the status?” instead of acting.

  • 🧠 Tribal knowledge: lost when people leave
  • 🔄 Manual status reports = outdated as soon as written
High process overhead
💸

Invisible waste

400+ team hours per year lost on coordination & chasing updates. No single source of truth across L&D projects.

€20k+ annual value at risk
🔗

Dependency blindness

Cross-project blockers (Kennishub → Marketing launch) surface only after delays hit critical path.

  • 📊 No automated dependency graph
  • 🚨 No early warning system
Risk multiplier
🧠

Karpathy knowledge stack

AI reads meeting notes, charters & KPIs → compiles a persistent wiki. Never forgets. Search, not index.

  • 📁 raw/ → wiki/ → outputs/
  • 🔍 hybrid search works at 200+ projects
Persistent brain
🔔

Proactive Monday alerts

Every week at 8:00 AM: Google Chat report lists bottlenecks, overloaded resources & blocked dependencies.

Same‑day detection

🚨 “Robin overloaded (120%) – document flow blocked”

⚙️

No lock‑in, open source

Claude (reasoning) + DeepSeek (batch) + 4iops dashboard. All self‑hosted, runs on your Google Drive & GitHub.

  • ✅ Data never leaves De Baak
  • ✅ No Microsoft vendor lock
100% Privacy-safe
👥

Human‑in‑the‑loop

AI suggests, townhall decides. Read‑only mode first, full trust after pilots. Improved decision quality & speed.

AI as assistant, not autopilot
💰

Minimal out‑of‑pocket

€250/year AI token budget (Claude + DeepSeek). Zero software licenses. 270 hours of internal time (already on payroll).

Less than €1/day
📈

8.000%+ ROI

400+ team hours saved annually → €20k+ value. Bottleneck detection from 3 days → same day. First L&D AI‑driven PMO.

  • ✨ 6‑week pilot, decision point
  • ✨ Risk: 82 hours + €55 worst case
High yield, low risk
📊

From 20 to 1.000 projects

20‑50 projects → full wiki compilation
200 projects → llmwiki hybrid search
500+ projects → multi‑agent swarm & exception routing

Modular architecture
⏱️

6‑week launch track

W1‑2 Setup & Karpathy brain
W3‑4 3 pilots (Kennishub, document flow, RACI matrix)
W5 First automated reports
W6 Townhall go/no‑go

Stop anytime after pilots