Personalisation Technology
Our Personalisation LLM fuses biometric signals, behaviour, and context to select the right content — privately and transparently.
Biometric Signals → Mind Map
Real-time signals like EEG, HRV, GSR and inferred mood blend with behavioural data (session context, prior viewing) and live context (weather, time of day). The map below shows how features flow into our Personalisation LLM, which selects content aligned to attention, comfort and preference.
- • EEG: focus/engagement features (alpha/theta balance)
- • HRV: relaxation / arousal proxy (RMSSD/SDNN)
- • GSR: moment-to-moment excitement
- • Mood: multi-signal inference (opt-in)
- • Behaviour: dwell, skips, completion, repeats
- • Context: time, weather, device, ambient light
Privacy & Opt-In
Personalised TV is optional, transparent and privacy-first. You stay in control.
Granular consent by signal type, with per-device controls.
Signal features derived locally where possible; raw data never leaves the device by default.
Ephemeral IDs, k-anonymity buckets, and DP noise for cohort metrics.
View and revoke history, export preferences, consent receipts.
Short TTL for session features; long-term only if explicitly enabled.
GDPR/CCPA aligned; DPIA and DSR processes in place.
Note: biometric features are derived representations — not raw medical signals — and are processed with consent and strict safeguards.
Personalisation LLM → Schedule Builder
The LLM turns preferences, context and constraints into a personalised schedule: format variety, pacing, ad-load targets, content diversity and safety rules — all tuned for each viewer.
- • Multi-objective ranking (engagement, novelty, diversity & fatigue-avoidance)
- • Session pacing (short/long form balance, ad breaks aligned to natural boundaries)
- • Safety / suitability filters (broadcaster policy & parental controls)
- • Exploration vs. comfort trade-off (context-aware discovery)
AI-Optimised HLS Playout
Adaptive bitrate ladders and LL-HLS tuned per viewer, device and network — reducing bandwidth and latency while preserving QoE.
AI selects per-title/per-device ladders, GOP alignment and keyframe cadence to minimize rebuffering.
Prefetch likely next segments at the edge; cut start-up delay and seek cost.
Live throughput estimation adjusts rendition switches long before stalls occur.
Edge hints + client telemetry feed back into real-time scheduling and caching.
The result: lower average bitrate for the same MOS, faster zapping, fewer rebuffers, and consistent latency in LL-HLS.
Bring personalisation to your platform
We integrate with your CMS, encoder and CDN — end to end, privacy-first.
Talk to our team