Looking for a job

Luis Ángel

@luisngel

Message

AI Engineer at Método Bebé Políglota who built WhatsApp CRM automations and evaluated an LLM reply classifier.

Colombia

Message

At Método Bebé Políglota, I built the operational messaging layer for a sales CRM using Python, Flask, and the WhatsApp Business API. It included an automatic retry for provider-blocked messages and a path for escalating replies that hadn't been picked up.

I also built and evaluated an LLM reply-intent classifier, compared its results with Claude-generated reference labels and a rules baseline, and created a reproducible script to audit VoIP billing. Alongside this work, I produced video content and analyzed performance metrics to optimize content and calls to action.

Experience

Work history, roles, and key accomplishments

Ingeniero de Inteligencia Artificial

Método Bebé Políglota

Jan 2024 - Sep 2026 (2 years 8 months)

https://github.com/onls294/reply-intent-classifier

Claude Haiku vs Claude Sonnet vs a rules baseline for routing WhatsApp sales replies. Evaluated on 164 real replies with bootstrap confidence intervals, macro-F1 and a costly-error analysis. Labels generated by Claude Opus and declared as such.

https://github.com/onls294/messaging-account-recovery-case-study

How a WhatsApp Business number's qua

Creador de Contenido

Método Bebé Políglota

Sep 2023 - Aug 2026 (2 years 11 months)

Worked as Creador de Contenido at Método Bebé Políglota.

Ingeniero de Inteligencia Artificial

Método Bebé Políglota

Aug 2023 - Aug 2026 (3 years)

CRM and AI automation
- Built the operational messaging layer for a sales CRM (Python/Flask, WhatsApp Business API): an automatic 26h retry for provider-blocked messages (recovering 23.9%, 16/67, within 72h) and a human-escalation path for replies no one had picked up.
- Built and evaluated an LLM reply-intent classifier: on 164 real, scrubbed replies, Claude Haiku matched Claude-generated referen