# eAlicia - Summary of Models and AI Workflow
eAlicia is an artificial intelligence platform developed by your team, Runcall Systems, since 2011,
focused primarily on call quality auditing and sales optimization.
It combines proprietary and modified open-source AI models to analyze phone interactions, measure performance KPIs,
and provide personalized improvement suggestions by agent, group, or service.

## 0. eAlicia Modules
- eAlicia specializes in measuring contact center quality.
- Customer satisfaction module for conducting surveys.
- Research & Education module for academic quality evaluations.
- eLearning module allows course uploads and online completion by students.

## 1. Types of Models in eAlicia

1. Audio Analysis Models:
   - Audio normalization and cleaning before analysis.
   - Temporal segmentation of the conversation.

2. Transcription Models (Speech-to-Text):
   - Convert audio to text with timestamps and speaker differentiation.
   - JSON output includes phrases and speech turns.

3. Language Models (LLMs):
   - Modified open-source and proprietary eAlicia models.
   - Main function: analyze transcribed text for KPI classification, reasons, and opportunities.
   - Fine-tuning and custom training per client and interaction type.

4. Analysis and Classification Models:
   - Quality KPI evaluation: friendly tone, issue resolution, compliance with LOPD protocol.
   - Detection of call reasons and sales opportunities.
   - Sentiment and emotion analysis.
   - Generation of recommendations and improvement suggestions per agent or group.

## 2. Processing Flow

1. Call Capture and Upload:
   - Audio in .wav or .mp3 format.
   - Stored on server with a processing URL.

2. Audio Preprocessing for Better Results:
   - Noise cleaning and normalization.
   - Diarization: separation of speakers and segmentation by turns.

3. Transcription:
   - Audio-to-text conversion.
   - JSON generation with timestamps and identified speakers.

4. Transcript and Audio Analysis:
   - LLMs classify KPIs, detect reasons, and analyze sentiment.
   - Internal models score performance and identify improvement opportunities.

5. Report Generation:
   - Dashboards with metrics by agent, campaign, and interaction type.
   - Historical reports and trend analysis using 14 years of interaction data.

6. Continuous Optimization:
   - Feedback from each call improves transcription and scoring accuracy.
   - Fine-tuned models are adjusted based on real outcomes and new scenarios.

## 3. eAlicia Priorities
- Transcription quality: foundation for text analysis accuracy.
- KPI and reason analysis: core business value.
- Model updating and feedback: maintaining well-tuned, client-adapted models.
- Scalability and efficiency: process large volumes of calls without precision loss.

## 4. Others
- eAlicia AI develops custom AI models for clients and universities: pattern detection, predictions, classification, etc.
- Tailored developments to meet specific client needs.