Transcript + Verbal + Visual
→ AI Model 116 → Engine 210 → Live Update
Not a black box. A board-auditable pipeline. Every signal has a claim, every score has a custody ledger. In 2022, before generative avatars, this was filed to make human-like adaptation adoptable — fast, affordable, and governed by CDR as foundation.
Calculated from positive, neutral, negative words. Word-count, not understanding.
Traditional conversational AI scored satisfaction by counting positive words, neutral words, negative words. Text-based sentiment and contextual analysis plus voice modulation. Suitable only for text and audio, not video. Unidirectional, sequential processing. Completely ignored emotional quotient while responding. It gave a dashboard after conversation — smile = happy label — but no action during conversation.
Intent + sequence + profile + tone + gesture. Context, not word-count.
I filed in 2022, before generative avatars, to move beyond satisfaction score. EI does not count positive words. It reads semantic context — why a person does what they do. User says "just browsing" but sequence shows comparing for 3 weeks at 2am — transcript alone misses intent. Add flat tone + pace drop + hand raise + eye drop = low conviction + high receptivity. That is semantic-based contextual analysis + objectified voice modulation + correlated body language analysis. Text, audio and video, with emotional quotient at core, bidirectional, live adaptation during streaming.
Emotion AI said: smile = happy label. That's a dashboard for analyst. No action for user. EI Model 116 says: conviction 42% LOW, attention 61% MID, receptivity 78% HIGH — will user disengage? Should we simplify? This is empathy-based learning per Claim 10, not emotion labeling.
Server AI Engine 210 takes trait scores + user profile + intent pattern and builds a prompt for Knowledge Graph Ontology. KG personalizes response — not just text, but tone, pace, visual empathy match. Voice clone mirrors calm, not excitement. Virtual avatar matches slouch with gentle posture, not stiff. This is Claims 15 and 17.
• Conviction 42% LOW
• Receptivity 78% HIGH
• Profile: first-time, anxious
• Pause, check understanding
• Voice: calm, slow
• Visual: open posture, eye contact
Emotion AI gave you dashboard after meeting: "user was disengaged at 14:32." Too late. EI gives you action at 14:32:01 — pause, simplify, suggest break. Tutor sees distracted → pause. Meeting bot sees low conviction → simplify language. Recruitment bot sees hand raise → pause interview. Yawn → suggest break. This is live-update multimodal content, not post-analysis.
CDR is foundation, EI is tool.
Two pillars, same building.
Data from EI is not stored as video. It is governed as custody ledgers F6,F7,F8 — Intent, Behaviour, Emotional. Zero retention, board-auditable, EU AI Act mapped.
| SIGNAL | WHAT IT CAPTURES | PATENT CLAIM | CDR LEDGER | EU AI ACT |
|---|---|---|---|---|
| Transcript | What they say + inferred intent | Claim 1 | F6 Intent Pattern | Transparency |
| Verbal Cue | Tone, pitch, pace shift | Claim 7 | F7 Behaviour Pattern | Art 5.1(f) • Workplace sensitivity |
| Visual Cue | Gesture, posture, hand raise, yawn | Claims 7,8 | F8 Emotional Pattern | Prohibition framework • Most dangerous |
| Trait Scoring | Conviction, attention, receptivity | Claim 10 | F7+F8 cross-ledger | No storage beyond session |
| Live Adaptation | Prompt + KG + live update | Claims 15,17 | Intent+Behaviour custody | Board-auditable • Zero retention |