Enterprise AI
Leadership, architecture and delivery at organisational scale.
Local AI first ≠Local AI only
Does every AI workload really need the cloud?
I experiment with real systems to find what should run Local, Hybrid or Cloud. I help teams make that decision with evidence.
Principal AI Engineer & AI Strategy LeadOpen-source systems
Where should AI run?
Start with the workload, not the model. Privacy, latency, control, capability and scale tell you whether Local, Hybrid or Cloud makes sense.
| Decision factor | Local AILocal-first lens | Hybrid | Cloud |
|---|---|---|---|
| PrivacyWhere data is processed | Local AIStrength: Stays on device | HybridSplit by policy | CloudProcessed remotely |
| Network & latencyConnectivity dependency | Local AIStrength: Works without network | HybridLocal fallback | CloudNetwork required |
| Runtime controlDeployment & model lifecycle | Local AIStrength: You control the runtime | HybridShared control | CloudProvider managed |
| Trade-offsWhere Local AI trades capacity for control | |||
| Model capabilityLargest model available | Local AIBound by device | HybridStrength: Routes by workload | CloudStrength: Frontier models |
| Elastic scaleBurst capacity | Local AIFixed by hardware | HybridStrength: Cloud overflow | CloudStrength: Elastic |
| Best fit | Local AIPrivacy, latency & control | HybridVariable workloads | CloudMaximum capability |
Not sure where your workload belongs?
Tell me what you're building and what constraints matter.
01 DECIDE 02 BUILD
Making local inference actually usable.
When Local or Hybrid makes sense, I build the runtime products need across desktop, mobile, and speech.
02 BUILD 03 TEST
Does it work in a real product?
Infrastructure only matters when a real application can depend on it.
Can sensitive document processing stay local?
WHAT THIS VALIDATES:
PII redactionHuman reviewPrivacy boundary
Can personal transactions be categorized on-device?
WHAT THIS VALIDATES:
On-device parsingDeterministic mathZero telemetry
Can meeting intelligence stay inside the room?
WHAT THIS VALIDATES:
Local audio ASRSpeaker contextOpt-in cloud
03 TEST 04 MEASURE
Performance Lab
Can this workload run locally?
Performance Lab benchmarks the real model on real devicesto turn evidence into a Local, Hybrid or Cloud decision.
Work in progress: representative benchmark dataset in active collection across target hardware.
Work in progress
Real evidence - my devices
Empirical benchmarkWhat matters
Speed
Memory
Stability
Quality
TTFT
Tokens/s
Peak RAM
Thermal
Task success
MacBook Pro M3 (36GB)Model: Nemotron Nano 4B Q4
0.24 s
34
3.1 GB
58 °C
91%
GoodFast, smooth generation and low thermals.
Samsung Galaxy A56 8GBModel: Qwen 3.5 2B Q4_K_M
0.62 s
13.5
1.9 GB
41 °C
76%
ReviewViable for targeted tasks; slower on complex context.
Deployment Synthesis
These results determine where the workload should run
LocalAll critical thresholds pass.
HybridLocal handles the main flow; harder cases need fallback.
CloudA hard requirement is not met locally yet.
Who I am
AI strategy. Hands-on engineering. Local AI.
I lead enterprise AI systems and build AI on real devices to understand what should run Local, Hybrid or Cloud.
Hands-on engineering
AI systems built and tested on Mac, Android and local infrastructure.
Public work
Open-source systems and technical writing through GitHub and stAI tuned.
AI Advisory
Not sure where your AIshould run?
I help teams make the right architecture decision and prove it on real systems.
Local•Hybrid•Cloud
01Decide
Where should it run?
I compare key factors to find the best fit for your workload.
- Privacy
- Latency
- Control
- Model capability
- Cost & scale
LocalHybridCloud
You getArchitecture decisionClear recommendation with rationale
02Design
How should it work?
I design the GenAI and runtime architecture around your needs.
- Model execution
- Memory & storage
- APIs & integrations
- Data boundaries
- Infrastructure options
Your dataAI runtimeOn-deviceCloud
(optional)
(optional)
You getReference architectureTechnical design and implementation plan
03Prove
Will it actually work?
I test on real hardware to validate performance, quality and limits before you commit.
- Performance
- Resource usage
- Privacy verification
- Quality & reliability
- Real-world constraints
Running inference...> Model: Qwen 3> Device: MacBook Pro> Tokens/s: 38.4> Memory: 4.1 GB> Status: ✓ Stable
You getValidation reportReal-world results and clear next steps
Local when it makes sense. Cloud when it doesn’t.Evidence decides.
- Independent advice
- No vendor lock-in
- Practical, unbiased, evidence-led