Agentic AI & LLMs
Multi-agent orchestration · MCP · Tool calling · LangGraph · Hybrid RAG · Voice agents · Real-time inference · vLLM · TensorRT-LLM.
Hi, my name is
Gerardo Barberena.
I ship agentic AI in production.
Senior AI/ML Engineer, 20+ years shipping production systems. Today I build real-time voice agents and agentic LLM platforms — Whisper ASR + neural TTS + tool-calling LLMs over hybrid RAG, served on NVIDIA H100 with sub-second time-to-first-token.
Check out my CV01. About Me
Hi! I'm Gerardo. I've been shipping software for 20+ years — riding the wave from early big-data on Hadoop and Spark, through the deep learning revolution (CNNs, RNNs, computer vision on edge), straight into today's frontier of multimodal LLM systems, agentic architectures and real-time voice AI on NVIDIA H100.
Today I lead AI/ML at Pragmatic / Neuranta, where we ship a real-time conversational AI agent that fuses Whisper ASR, neural TTS and tool-calling LLMs over a hybrid RAG layer backed by Milvus — orchestrated on Kubernetes and served on H100 GPUs with sub-second time-to-first-token. Before that: AI-driven cybersecurity, real-time POS analytics across Alsea brands (Starbucks, Domino's), and large-scale SCADA telemetry over 1,500 industrial gas wells.
The thread across two decades stays the same: ship AI that runs in production at scale — not in slides.
Frontier stack I'm building with:
- Agentic AI · MCP · Tool calling
- Voice agents · Whisper · TTS
- Hybrid RAG · Milvus · pgvector
- vLLM · TensorRT-LLM · H100
- LangGraph · MLflow · Airflow
- PyTorch · CUDA · Kubernetes
02. Where I've Worked
Applied ML Engineer @ PRAGMATIC / Neuranta
Nov 2022 — Present
- Multimodal conversational system combining ASR (Whisper), TTS and LLMs with RAG.
- Milvus vector DB for semantic retrieval and dynamic call-flow context.
- Kubernetes microservices for high availability and horizontal scale.
- GPU-accelerated pipelines on NVIDIA H100 for low-latency ASR + LLM generation.
Machine Learning Developer @ Xtrmedata / PRAGMATIC
Mar 2021 — Nov 2022
- CNN/RNN deep learning models for real-time person counting on RTSP feeds.
- User-behavior ML, SIEM correlation and automated incident response.
- Spark-based ingestion + TensorFlow for advanced threat detection.
- Custom monitoring agents feeding Spark pipelines for large-scale analysis.
Machine Learning Developer @ Alsea (Starbucks · Domino's)
Nov 2014 — Dec 2020
- Real-time anomaly detection and camera-based analytics in production.
- End-to-end platform monitoring transactions, electric meters and network traffic.
- Unified POS data across Alsea brands to detect anomalies and optimize sales.
- AWS (EC2, EMR) ETL pipelines integrating camera feeds and meter data.
Intermediate Software Engineer @ IT Guardian — Financial Analytics
Dec 2013 — Sep 2014
- ML algorithms (A/B testing, association rules, gradient boosting) on trading patterns.
- Predictive models for market trends, optimizing automated trading desks.
- Spark-on-Hadoop ingestion, feature engineering and model training.
Developer @ PEMEX — SCADA
Jan 2012 — Sep 2013
- Deep learning on real-time sensor data across 1,500 gas wells.
- MongoDB map-reduce for geolocation-based anomaly detection.
- Remote command execution and real-time dashboards for maintenance.
Developer @ SIAVE-Aduanas
Apr 2009 — Oct 2011
- Rules engine with ML-based heuristic predictions for customs data.
- Real-time alerts flagging high-risk shipments from cargo + historical patterns.
- Integration with external government databases for compliance checks.
03. Skills & Stack
Deep Learning & Foundation Models
PyTorch · TensorFlow · Transformers · LoRA / QLoRA fine-tuning · Diffusion · CNN / RNN · OpenCV · Whisper · Speculative decoding.
Data, Vectors & Big Data
Milvus · pgvector · Qdrant · Spark · Hadoop · Kafka · Pandas · NumPy · Scikit-learn · SQL.
GPU Infra & Cloud
NVIDIA H100 · A100 · Jetson Orin · Triton · KServe · Kubernetes · Docker · AWS (EC2, EMR, SageMaker) · Azure · GCP.
MLOps & LLMOps
MLflow · Airflow · Kubeflow · LangSmith · Langfuse · RAGAS evals · Jenkins · CI/CD · Production eval pipelines.
Languages
Python · Java · C / C++ · SQL · C# · Visual Basic.
04. What's Next?
Get In Touch
Open to senior AI/ML roles, LLM platform engineering and agentic AI consulting. The fastest way to reach me is WhatsApp — usually reply within a few hours.
WhatsApp +52 55 7853 9814