Computer Vision for Broadcast
Detection, segmentation, tracking — tuned for broadcast reality.
- Mask-level segmentation
- Occlusion & camera motion
- Performance profiling
AI • Computer Vision • Live Broadcast Systems
AI that survives live production.
I build computer vision and real-time software for sports broadcast: sponsor replacement, graphics automation, and timing/data pipelines. Built for messy footage, hard deadlines, and zero tolerance for failure.
Segmentation (Mask-level)
Tracking
GPU pipelines
Broadcast integrations
Now
Open to UK roles + consulting projects
Strength
Production-first engineering
Focus
Sports broadcast AI
Edge
Real-time + operator workflows
Featured work
One strong project > ten vague ones. This is the one.
AdSwap: before / after
AlterVision / AdSwap
Sponsor board detection & replacement
Computer vision pipeline that detects pitch-side boards, generates clean masks, and swaps creatives while keeping perspective, motion, and occlusions stable.
• Mask-level quality (not just boxes)
• Built for broadcast footage (compression, camera movement, clutter)
• Designed to integrate into live/near-live workflows
Services
Clear offers. Clear outcomes.
Real-time Video Pipelines
NDI/SRT/file pipelines that hit FPS targets without falling apart.
- GPU decode/resize strategies
- Latency control
- Observability (logs/metrics)
Timing & Data Systems
Operator-first UIs, reliable storage, clean feeds for graphics/web.
- SQL-backed reliability
- Operator workflow design
- Export APIs for broadcast
Why teams trust this
This is the “I’ve been on-site when things go wrong” section.
Production mindset
I design for failure modes: flaky networks, bad lighting, operator stress, and last-minute changes. If it can break on show day, I assume it will — and build around it.
Bridge: AI ↔ Broadcast
Most AI demos die at integration. I don’t stop at “model works” — I ship the full pipeline: ingest → process → validate → publish → monitor.
Fast iteration
Prototype quickly, measure everything, then harden. You’ll see progress early, not after 3 months of “research”.
Operator-first UX
Live ops is a human system. If the UI isn’t obvious at 200 bpm heart rate, it’s not done.
ML education
Formal training that actually feeds into practical systems.
Imperial College London — Professional Certificate (ML & AI)
Current coursework and applied projects across supervised/unsupervised learning, optimisation, interpretability, and modern deep learning.
Machine Learning and Artificial Intelligence
Open to roles
Straight to the point: what I’m looking for.
Target roles
- Computer Vision Engineer (sports / broadcast)
- ML Engineer (applied, production pipelines)
- Software Engineer (video, real-time, data)
- Product-minded Engineering roles in media tech
What I want to build
- Vision systems that integrate into live workflows
- High-quality masking/segmentation for graphics & overlays
- Scalable pipelines with cost control and monitoring
What you’ll get in 30 days
- A working demo in your footage/domain
- Measured performance (FPS/latency) + next-step plan
- Integration notes (APIs, deployment, monitoring)
Let’s build something that works on show day.
Send the problem, the deadline, the footage type (live/recorded), and what success means (latency, quality, cost).
Quick notes
LocationReading, UK
WorkRoles + consulting
FocusSports broadcast AI