Table of Contents

Edge AI Business Roadmap

2-Year Plan: Skills → Consulting → Product → Passive Income

---

Strategic Overview

Build a high-value consulting practice around Edge AI while developing a productised offering that generates passive income. Four phases, starting with a focused 6–8 week skills sprint before any client outreach.

Your unfair advantages: Cloud/K8s architecture, MLOps experience, deep learning knowledge, existing paying client, own hardware infrastructure, Estonia/EU market access.

---

PHASE 0 — Skills Sprint

Weeks 1–8 | Build before you sell

Do not start marketing yourself until this phase is complete. One working demo is worth more than any CV.

---

Week 1–2: Hardware & Baseline Inference

• Buy Raspberry Pi 5 or Jetson Orin Nano Super (250$)
• Source a USB or CSI camera module
• Install base OS, configure SSH and remote access
• Deploy a pre-trained YOLO model, run live inference on camera feed
• Verify hardware acceleration (GPU/NPU) where available
• Document setup process as you go (becomes your first case study)

---

Week 3–4: Model Optimisation & Quantization

• Set up RunPod or Vast.ai account (pay-as-you-go, don't buy GPU yet)
• Download a base YOLO or Florence 2 model
• Run INT8 / INT4 quantization — benchmark accuracy vs speed tradeoff
• Export optimised model (ONNX / TFLite / NCNN for target device)
• Deploy quantised model to edge device — record FPS before vs after
• Note your real gaps — this is where you find what to study next

---

Week 5–6: Monitoring Stack & Remote Management

• Set up Grafana + Uptime Kuma on your existing server node
• Stream edge device telemetry to Grafana (CPU, temp, inference latency, FPS)
• Build a simple OTA model update mechanism (pull new weights remotely)
• Configure alerting for device downtime or inference degradation
• This stack becomes your managed service product — treat it seriously

---

Week 7–8: Case Study & Outreach Prep

• Write up the full build as a case study with benchmarks and screenshots
• Pick ONE target vertical: manufacturing, agriculture, or retail (manufacturing = highest budget in Nordics)
• Write a one-page positioning statement: problem / who / cost
• Record a short demo video or build live demo capability
• Update LinkedIn profile with new positioning (see Phase 1 for exact wording)
• Identify 10 target companies in your vertical in Estonia / Nordics
• Draft outreach message templates — offer a free 2hr edge AI assessment

---

PHASE 1 — Consult to Fund

Months 2–6 | Generate revenue, deepen skills

Use existing cloud/MLOps skills positioned toward edge AI evaluation. You are the bridge between cloud and edge — that is a real and paid consulting need right now.

---

Positioning & Sales

• Use title: "Cloud & MLOps Architect — Edge AI Infrastructure"
• Do NOT claim "Edge AI Expert" yet — you are the person who helps companies decide whether and how to go to edge
• Target: companies exploring edge AI who need architecture guidance before committing
• Rate: €400–800/day — your cloud architecture background justifies this immediately
• Aim for 2–3 consulting days/week to protect time for product development
• Send outreach to the 10 companies identified in Week 8
• Attend 1–2 relevant meetups or events in Estonia / Nordics (in-person beats LinkedIn cold outreach)

---

Infrastructure

• Buy second cheap server node €400–600 for redundancy on existing client
• Set up Coolify or Dokploy across both nodes for professional deployment management
• Run Portainer, Grafana, Minio, model registry on new node
• This is your managed service backbone — not a data centre, a professional platform
• Do NOT buy a GPU yet — use RunPod for all training/quantization work

---

Skills to Build in Parallel

• K3s / KubeEdge — Kubernetes at the edge (direct transfer from your existing K8s skills)
• MQTT and edge messaging protocols
• OTA update patterns for embedded/edge systems
• Basic security hardening for edge devices (future managed service upsell)

---

Revenue Targets

• €3,000–6,000/month consulting revenue by end of Month 6
• First managed service client on recurring fee — even €200–500/month
• Hardware node paid off through company expenses

---

PHASE 2 — Build the Product

Months 4–10 | Productise the prototype

Consulting income funds this. Run in parallel with Phase 1 — don't wait for Phase 1 to finish.

---

Confirm Your Vertical

• Manufacturing defect detection — highest budget, clear ROI, repeat hardware orders
• Agricultural monitoring — EU funding available, low competition, growing fast
• Retail footfall / shelf analytics — easier demo, shorter sales cycle
• Recommendation: manufacturing if you want highest value contracts in Estonia/Nordics

---

Product Build

• Use Florence 2 to auto-label training data for your vertical (eliminates manual annotation cost)
• Fine-tune YOLO on vertical-specific dataset using RunPod GPU time
• Package as edge device + software bundle (hardware + annual licence model)
• Build simple web dashboard for client-facing monitoring and reporting
• OTA model update pipeline — clients always receive latest model version
• Define pricing: hardware margin + €200–500/device/year licence + managed service fee

---

When to Buy a GPU

• Buy a used RTX 3090 (€600–900) once you have your first product client
• 24GB VRAM handles YOLO training, Florence 2, quantization of models up to 13B
• Secondhand market is well stocked right now — don't overpay for new
• Add to company expenses — tax deductible capital asset

---

Managed Service Stack (Passive Income Foundation)

• Model deployment & OTA updates → charge €200–500/device/year
• Uptime monitoring → €100–300/month per client
• Security scanning / threat detection → €300–800/month per client
• Compliance reporting (GDPR, ISO) → €500–1,500/month per client
• A single manufacturing client with 10 edge devices = €2,000–4,000/month recurring

---

PHASE 3 — Scale Consulting + Passive Income

Months 10–24 | Leverage and recurring revenue

---

Consulting Rate Escalation

• You now have a live product and domain expertise — justify €150–300/hour rates
• Productised consulting: fixed-price edge AI readiness assessments (€2,000–5,000 flat fee)
• Offer managed deployment packages on top of assessments
• Target larger clients in manufacturing / logistics / agriculture across Nordics and EU

---

Infrastructure Expansion (Only If Revenue Justifies It)

• Reinvest managed service revenue into additional hardware — never buy ahead of revenue
• Each new hardware node should be funded by a new managed service contract
• Consider colocation if bandwidth becomes a constraint (you'll know when)

---

Passive Income Stack

• Product licence fees (recurring annual per device)
• Managed service fees (monthly per client)
• Model update subscription (clients pay for continued model improvement)
• Target: €5,000–15,000/month recurring by Month 24 alongside consulting

---

Summary Timeline

Phase	Timeline	Primary Goal	Revenue Target
Phase 0 — Sprint	Weeks 1–8	Build real edge AI skills and demo	€0 (invest time)
Phase 1 — Consult	Months 2–6	Paid consulting, fund product build	€3–6k/month
Phase 2 — Build	Months 4–10	First product client, recurring fees begin	€1–3k/month recurring
Phase 3 — Scale	Months 10–24	High-rate consulting + growing passive income	€5–15k/month recurring

---

Key Principles to Keep in Mind

• Build before you sell — one working demo beats a polished pitch every time
• Consult to learn — every client conversation tells you what the product needs to be
• Hardware only when clients justify it — you've already proven this model works
• Own the infrastructure — renting cloud long-term destroys your margin advantage
• Pick one vertical and go deep — generalist edge AI is a commodity; vertical expertise is not