Company Description

Meta‑Flux is a tech‑bio company building an AI‑driven drug‑development platform that fuses multi‑omics data (metabolomics, proteomics, transcriptomics) with graph neural networks, knowledge graphs and agentic workflows to accelerate therapeutic development. Our stack pairs a user‑facing app with a dedicated Graph Service running GPU‑accelerated analytics (PyTorch Geometric, cuDF/cuGraph) on AWS, and leverages Bedrock for LLMs within orchestrated, tool‑using agents (LangGraph preferred).

Role Description

You’ll be one of two AI engineers driving our next‑gen modeling initiative, from research prototype to production deployment across a modern AWS stack. You’ll collaborate tightly with our bio and data teams, own the training/evaluation pipeline, and help shape engineering standards around reliability, security, and compliance.

Lead our next‑gen multi‑omics modeling initiative: a graph‑aware transformer that fuses omics signals with sample‑level metadata to learn shared representations that power biomarker and target discovery. You’ll prototype core blocks, define training/eval strategy, and take models from research to production.

What You'll do

• Prototype graph‑conditioned transformer components and training loops.
• Define multi‑omic encoding/tokenization for aligned samples across modalities (e.g., metabolomics, proteomics, transcriptomics).
• Run self‑supervised objectives (masked modeling, contrastive alignment, cross‑modality prediction) to learn robust, reusable representations.
• Own scalable training & evaluation on GPUs (throughput, cost, profiling, mixed precision, checkpointing, reproducibility).
• Drive ablations, stress tests, and robustness/shift evaluations; document findings and convert them into product wins.
• Build agentic workflows (LangGraph) that orchestrate data prep, model calls (Bedrock), and validation steps end‑to‑end.
• Partner with bio specialists to align modeling assumptions with experimental reality and prior knowledge (pathways/ontologies/graphs).
• Productionize models behind clean APIs and batch pipelines; instrument with logging, tracing, and experiment tracking.
• Contribute to security‑minded engineering and SOC 1 / SOC 2 readiness (data handling, auditability, least‑privilege, reproducibility).

Qualifications

• 5+ years of hands‑on ML/AI engineering (or equivalent impact), including shipping models to production.
• Deep proficiency in Python and PyTorch (or TensorFlow) with strong software engineering fundamentals.
• Strong background in transformers and at least one of: GNNs, contrastive/self‑supervised learning, representation learning.
• Comfortable with GPU training stacks: CUDA, PyTorch AMP, profiling, memory/perf debugging.
• Familiarity with graph/data tooling: PyTorch Geometric, cuDF/cuGraph, NetworkX‑equivalents.
• Practical AWS experience with at least some of: EC2, SageMaker, Bedrock, RDS (or similar managed DBs), and a managed/hosted graph database.
• Experience building reliable data/ML pipelines and evaluations at scale.

Nice to have

• Background working with omics datasets or adjacent bio/healthcare data; comfort reading domain literature.
• Scientific AI exposure: scGPT, AlphaFold, Boltz, ESM‑3, or related bio/chem models.
• LangGraph or LangChain experience building agentic systems; streaming, tools, and multi‑step orchestration.
• Knowledge graph experience (schema design, ingestion, enrichment, query) and pathway/ontology integration.
• Security/compliance minded engineering; prior exposure to SOC 1/SOC 2 programs.
• Experiment tracking and observability (e.g., MLFlow/Weights & Biases), structured logging, and testable research code.

Ways of working

• Hybrid in Ireland. Focus time at home; periodic in‑person collaboration.
• Small, senior‑leaning team; high ownership; pragmatic research‑to‑prod cadence.
• We value clear writing, thoughtful code reviews, and reproducible experiments.