Interaction analytics & AI engineering
Turn customer conversations into advisor insights.
Connected interaction capture, transcription, AI insights and business reporting.
Team experience / Entel. Implementation led by Mathyu Cardozo
Project
SinfonIA / Entel
Industry
Telecommunications
Development
5 months of development
Stack
WhisperX, AWS SQS + Lambda, OpenAI APIs, Python / FastAPI, Next.js / React, New Relic
Evidence
Tech Lead-reported · 100K+ calls processed daily
Mathyu Cardozo · AI Engineer Lead
In production, with monthly maintenance and new functionality on demand.
The challenge
Customer information remained inside recordings and messages.
What we built
An internal platform connected capture, transcription, personal-data treatment, AI insights and reporting.
What changed
100K+ calls daily and approximately 75% lower operating cost, reported by the Tech Lead.
Context and problem
Customer information inside recordings and messages needed a connected route into advisor evaluation and business reporting.
What we built
Implementation at a glance
Capture and transcribe interactions using the reported AWS processing foundation.
Apply personal-data treatment and generate AI insights.
Connect the analysis with evaluation, reporting and operating observability.
Implementation details
Useful information remained inside recordings and messages. Entel needed an internal processing path that could turn those interactions into usable evaluation and reporting while reducing dependence on its previous analytics provider. The challenge covered the complete handoff from source interaction to information a business team could use.
Build a connected processing path
The voice workflow connected capture, transcription with WhisperX, personal-data treatment, AI insights and reporting. SQS and Lambda supported the processing foundation.
Connect analysis to business output
OpenAI APIs participated in the insight engine, while the application stack included Python/FastAPI and Next.js/React. WhatsApp and internal channels expanded the interaction sources beyond voice.
Support changes to an operating product
The reported delivery practices included development, staging and main branches, a weekly release window, E2E checks before and after deployment, and observability through New Relic.
Results and evidence
The Tech Lead reports more than 100,000 calls processed per day and approximately 75% lower operating cost than the previous provider. The source does not disclose a comparison period or normalized cost base. These figures do not establish annual savings or isolate the contribution of one technology.
Relevance and PDF contents
Where this pattern applies
This pattern is relevant when customer information remains inside conversations and the team needs a repeatable route to a decision or report. Begin with one interaction source, its processing stages and a defined evaluation output. The PDF explains the functional architecture and reported operating practices in more detail.
Inside the full PDF
Explore the processing architecture, delivery practices and reported results behind SinfonIA.
Five-stage interaction-to-reporting workflow
Processing, insight and application responsibilities
Release checks and operating observability
Interaction-analytics readiness worksheet
Free PDF · 8 pages · Includes a readiness worksheet.
On this page
Follow the pipeline from captured interactions to transcription, AI insights and reporting. Use the worksheet to assess a similar analytics workflow.
Next case