Unai Garay Maestre — Senior AI Engineer | Agentic RAG & LLM Systems

About me

Most of my work is agentic RAG and complex LLM systems — retrieval pipelines, agents that call real tools, and the classical ML and NLP that still beat a prompt for clustering, ranking and extraction. Neural search with RAG and agents over unstructured medical data at Sorcero, an LLM-driven ticketing pipeline at Tekever, computer vision at scale before that. Lately that has extended into production voice agents too.

What I actually care about is the unglamorous half — the evals, guardrails and infrastructure that decide whether a demo survives contact with real users. I've shipped that half at a startup, a scale-up and a consultancy, and I like the teams I work with to be better for my having been there.

Away from the keyboard I'm playing music or outdoors — bungee jumping, climbing, paddle boarding. I also shoot a lot of photography, which lives on a separate Instagram.

Read the résumé See the projects

Technical skills

AI & machine learning

RAG
Agents
LLMs
LLM evals
NLP
Neural search
Classical ML
Computer vision
PyTorch
Transformers
Voice AI

Engineering

Python
FastAPI
Docker
AWS
Google Cloud
Airflow
React / TypeScript
C# / .NET
Angular
React Native

Research

Variational autoencoders
Spiking neural networks
Ontology integration
Feature engineering

Work experience

Current role

Inulti

Senior AI Engineer June 2026 – Present

Building production real-time voice AI agents for inbound sales calls (LiveKit + SIP telephony, Deepgram STT, OpenAI LLM, ElevenLabs TTS) with sub-second turn latency, warm transfer to human agents, and PCI-safe payment capture over voice and phone keypad.
Delivered an LLM retention platform on AWS end to end: AI-drafted SMS replies with safety guardrails and human-in-the-loop review, a CRM over ClickHouse, and LLM-as-judge plus synthetic-call eval harnesses gating every prompt and model change.

Python
Voice AI
LiveKit
LLM Agents
FastAPI
React/TypeScript
AWS
Docker

Tekever

Senior LLM Engineer / Data Scientist April 2025 – June 2026

Engineered an end-to-end data pipeline that turns Signal chats into Jira tickets automatically, clustering conversations by semantic similarity and temporal proximity and using LLMs to generate the structured ticket data.

Python
Airflow
Docker
LLMs
Google Cloud Platform

Sorcero

Senior AI Engineer March 2020 – 2025

Led AI R&D for a natural language understanding platform for unstructured medical data: generative theme tagging, NER, summarization, recommendation, and a custom neural search engine with retrieval-augmented generation and agents. The work behind three US patent filings.

Python
NLP
PyTorch
spaCy
Transformers
FastAPI
Docker
Google Cloud Platform

Accenture

Data Scientist May 2019 – March 2020

Built a tool that applies data analysis, clustering, topic modelling and other NLP methods to ticketing-system problem descriptions to speed up ticket resolution.
Built an NLP tagging service that infers people's skills and interests for an employee selection tool.

Python
NLP
Scikit-Learn
NLTK
spaCy
Plotly/Matplotlib

Smartloto S.L.

AI Engineer July 2018 – April 2019

Built an end-to-end image validation system for lottery tickets using deep learning at scale on Google Cloud Platform, with on-device inference on Android and iOS.