EngineMind EFT: Detecting Emotion in Text at Machine Scale

Most sentiment analysis treats emotions like a binary toggle — positive or negative. EngineMind's Emotional Framework Translator maps text to Plutchik's eight primary emotion vectors, giving you nuance that scalar sentiment scores miss entirely.

TL;DR

EngineMind EFT maps text to Plutchik's eight primary emotion vectors — joy, sadness, anger, fear, surprise, anticipation, trust, and disgust — at machine scale. Useful for agentic pipelines that need to understand not just what a user said, but how they felt when they said it.

The 10-Second Pitch

Plutchik-based emotion vectors — eight primary emotions mapped from text, not just a positivity score
Fine-grained intensity scores — each emotion has a 0-1 confidence score, not just a dominant label
Designed for LLM output — works well as a downstream classifier on agent-generated content
Multilingual — supports English, Spanish, Mandarin, and French
Streaming API — process individual utterances or batch-analyse conversational logs

Setup in 2 Steps

1. Get credentials. Register at enginemind.ai. Set EFT_API_KEY in your environment.

2. Configure the skill:

mrt install enginemind-eft

Prompt to test it:

Analyse this customer support transcript and give me the dominant emotion per speaker, with intensity scores:
Customer: I have been waiting for my order for three weeks and I am absolutely furious.
Agent: I completely understand your frustration and I am looking into this right now.

How It Actually Works

EngineMind EFT runs a fine-tuned transformer model trained on human-annotated emotional speech corpora. It classifies each utterance against Plutchik's wheel. The model returns eight float scores summing to approximately 1.0 per utterance, with the highest-scoring pair indicating the dominant emotional posture.

Why This Matters for Agents

Standard sentiment analysis is built for review stars and NPS scores. Agentic AI needs more — it needs to know when a user is mask-slipping from frustration to anger. A scalar sentiment score of 0.3 (slightly negative) does not tell you whether the user is fear[0.4] + anger[0.3] or sadness[0.5] + anger[0.2]. Those require very different agent responses.

Pros / Cons

Pros	Cons
Emotionally granular vs. scalar sentiment	Requires API key / self-hosted model
Plutchik vectors are psychologically grounded	Intensity scores are relative, not absolute
Works as downstream classifier on LLM output	Lower accuracy on sarcasm and irony

Verdict

If you are building customer-facing agents and you are not doing emotional classification, you are flying half-blind. Scalar sentiment is useful for dashboards; EFT is useful for agents that need to modulate their responses. Worth adding to any serious conversational AI pipeline.