# Daniel Romero

> AI Engineer with 25 years in tech, working across the full stack of intelligent
> systems: from GPU kernel optimization with Triton, through production RAG over
> 70M+ documents, to end-to-end agentic workflows. Based in Fortaleza, Brazil,
> working remotely.

Daniel Romero worked inside OpenAI, on a project the company acquired. His career
runs from Linux systems administration in 2000, through software development and
site reliability engineering, to AI engineering today. He takes LLM systems from
research into production, and is currently building and training a humanoid robot.

This file describes <https://infoslack.pro/>, a single-page site. All of the
content below is on that one page.

## Contact

- Email: <infoslack@gmail.com>
- GitHub: <https://github.com/infoslack>
- LinkedIn: <https://www.linkedin.com/in/infoslack>
- X: <https://x.com/infoslack> (@infoslack)
- dev.to: <https://dev.to/infoslack>
- YouTube: <https://www.youtube.com/@daniel_romero> (videos about AI, in Brazilian Portuguese)
- Website: <https://infoslack.pro/>

## Where he works in the stack

Five layers, from the highest level of abstraction down to the hardware.

- **Agents**: end-to-end agentic workflows, covering planning, tool calls and
  multi-step execution against real systems. LangChain, LlamaIndex, DSPy.
- **Retrieval**: hybrid search at 70M+ documents, combining dense, sparse and
  late-interaction embeddings with re-ranking. Qdrant, HuggingFace, OpenAI.
- **Serving**: ingestion, embedding generation and query APIs. FastAPI, Docker,
  Kubernetes.
- **Model runtime**: mixed-precision training in FP16 and BF16, plus INT8
  quantization. PyTorch, CUDA.
- **Kernels and silicon**: high-performance Triton kernels shipped as PyTorch
  extensions, written after profiling the GPU. Triton, Nsight, PyTorch Profiler.

## Experience

- **Stealth AI startup (acquired by OpenAI)**, AI Engineer. Remote, USA.
  07/2025 - 02/2026. Built large-scale LLM workloads with pipelines integrated
  directly into OpenAI production systems, and continued on the project inside
  OpenAI after the acquisition. Wrote high-performance Triton kernels shipped as
  PyTorch extensions, cutting GPU inference latency at the kernel level. Profiled
  GPU bottlenecks with Nsight and the PyTorch Profiler, then applied
  mixed-precision training (FP16/BF16) and INT8 quantization.
  Stack: Python, PyTorch, Triton, CUDA, LangChain, LlamaIndex, Qdrant, FastAPI.
- **Doximity**, Senior Software Engineer / AI. Remote, USA. 07/2024 - 07/2025.
  Designed and shipped a production RAG platform for semantic search across 70M+
  medical documents. Combined dense, sparse and late-interaction embeddings with
  re-ranking, beating the keyword-search baseline on retrieval precision. Built
  the FastAPI backend covering embedding generation, document ingestion and
  vector search.
  Stack: Python, FastAPI, Qdrant, HuggingFace, OpenAI, LangChain, Docker,
  Kubernetes.
- **Qdrant**, ML Engineer / AI. Remote, Germany. 01/2024 - 07/2024. Wrote the
  tutorials and documentation for semantic search, vector search and RAG.
  Contributed framework integrations for DSPy, LangChain and LlamaIndex. Recorded
  a tutorial series for the Qdrant YouTube channel.
  Stack: Python, LangChain, LlamaIndex, DSPy, Qdrant, OpenAI, HuggingFace.
- **DNSFilter**, Platform Engineer / ML. Remote, USA. 06/2022 - 12/2023. Cut
  deployment time 87%, from 5 minutes to 40 seconds, moving Rails services off
  bare metal onto Kubernetes (EKS). Closed monitoring blind spots with continuous
  alerting and observability. Contributed to the deployment pipeline the data
  science team used to ship ML models to production.
  Stack: Terraform, Ansible, Docker, Kubernetes (EKS), Loki, Grafana, Datadog,
  eBPF, PyTorch.
- **Henku**, Data Scientist / ML. Remote, Estonia. 04/2021 - 06/2022. Cut model
  training and deployment 90%, from 30 minutes to 3, replacing ad-hoc notebooks
  with a continuous training pipeline serving 10K+ active users in Spain. Took a
  food-nutrition computer vision model from scratch to production with a CNN in
  PyTorch. Automated the model lifecycle with Airflow, Kubeflow and Weights &
  Biases.
  Stack: Python, PyTorch, TensorFlow, Scikit-Learn, Airflow, Kubeflow, Weights &
  Biases.
- **Elastic**, Senior Site Reliability Engineer. Remote, USA. 09/2018 - 04/2021.
  Held 99.95-99.99% uptime across tens of thousands of Elasticsearch clusters on
  AWS, Azure and GCP, leading on-call in a multi-region, multi-cloud environment.
  Contributed to a Golang CLI for cluster maintenance that took roughly 8 hours of
  manual work per week off the SRE team. Deployed new regional infrastructure for
  Elastic Cloud.
  Stack: Golang, Python, Ansible, Terraform, Kubernetes, AWS, Azure, GCP,
  Elasticsearch.

Earlier: DevOps Engineer (2015 - 2018), Software Developer (2007 - 2015, PHP,
Java, Ruby on Rails), Sysadmin (2000 - 2007, Linux, iptables, networks, Bash,
Perl).

## Physical AI

Daniel is building and training his own humanoid robot. It is the same work he
does with language models, pointed at hardware, with physics as the feedback
loop. Work in progress, written up as he goes:

- [Building my humanoid robot](https://dev.to/infoslack/building-my-humanoid-robot-pdg)
- [Building a physical API](https://dev.to/infoslack/building-a-physical-api-3488)

## Optional

- [Full CV as PDF](https://infoslack.pro/assets/Daniel-Romero.pdf): the same
  history in resume form.
- [Qdrant tutorial playlist](https://www.youtube.com/playlist?list=PL9IXkWSmb369tDKcdBf-NxbjcCHcx-9Al):
  the video series he recorded on vector search and RAG, in English.
- [Daniel Romero on YouTube](https://www.youtube.com/@daniel_romero): his own
  channel, covering AI in Brazilian Portuguese.

Note on language: this site, the CV and the dev.to posts are in English. The
personal YouTube channel is in Brazilian Portuguese. Daniel speaks both.