This site is under development. For the current active site, go to bcbhub.csic.es.

Researcher

Francisco Miguel Pérez Canales

Biography

I build AI systems that survive contact with reality: large-scale data, evolving ground truth, and the gap between a notebook that works once and a traceable platform that runs every day. End-to-end: research, infra, backend, and the frontend that puts it in front of users.

Ten years across applied ML and computational science, now continuing in my PhD at Universidad de Sevilla. I own the full research-to-production pipeline: framing the problem, modelling the data, training, evaluation, deployment, monitoring, authentication, the API + web + agent layers, and the React / Next.js UI on top.

Recent highlights

  • PROTEA, a production-grade ML platform: protein language model inference at scale (ESM2, ESMC, ProstT5, Ankh), retrieval-augmented prediction over 500k+ embeddings (FAISS + pgvector halfvec), FastAPI + Next.js + RabbitMQ orchestration, Grafana / Loki / Prometheus observability, OAuth / LDAP auth, Dockerised CI/CD on GitHub Actions.
  • agent-farm (OSS), a multi-agent orchestration framework: conductor session spawning 10+ specialised workers, MCP integration, per-agent permission scoping, SQLite state, token-cost-aware mixed-model dispatch.
  • #19 of 2,186 teams in CAFA 6, Kaggle Gold Medal. International benchmark for automated protein function prediction. Per-aspect calibrated ensemble of four protein language models.
  • Publications: co-author, Communications Biology 2025 (Nature portfolio); first author, Methods in Molecular Biology 2025 (Springer); co-author, "FANTASIA suite" submitted to NAR Genomics and Bioinformatics, May 2026.

Stack

  • Languages: Python · TypeScript · JavaScript · Bash · C/C++
  • ML / DL: PyTorch / Lightning · Hugging Face Transformers · LLMs · Multi-Agent Systems · MCP · RAG
  • Data / Retrieval: FAISS · pgvector · PostgreSQL · SQLAlchemy · RabbitMQ · MongoDB
  • Backend / API: FastAPI · Node.js · Express · Django · REST · GraphQL
  • Frontend: React · Next.js · Angular · Tailwind CSS · next-intl
  • Infra / HPC: Docker · Kubernetes · SLURM · Apptainer / Singularity · Nextflow · GitHub Actions · AWS
  • Observability: Grafana · Loki · Prometheus
  • Auth: OAuth · LDAP · SAML · SOAP

Profile

Institution US
Position Bioinformatician, MLOps Engineer and AI Engineer
Home page https://github.com/frapercan
Contact email fmpercan@upo.es
ORCID

BCB Hub

BCB Committee Not a member
BCB Tools
BCB Services

Research

Research topics

Artificial IntelligenceMachine Learning in BiologyBioinformatics Software and ToolsLarge language modelsFunctional annotationFunction predictionBiological DatabasesHigh-performance ComputingContainers ComputingWeb-serviceWorkflow managersProteomics
View all (2 more)
Deep Learning in BiologyCloud Computing

Research lines

None

Publications 0 entries

Suggest changes

Spotted something outdated or wrong here? Let us know below and we'll take care of it.

← Return to Home