LLM · Manufacturing
AI Quotation Assistant
A fine-tuned language model that reads engineering requirement files and produces structured quotations in minutes instead of days.
~ case study on request
Independent Research Studio · Est. 2025
Applied intelligence, engineered with the
rigor of research.
Phoenix Labs is a solo studio building production AI, machine-learning systems and secure cloud infrastructure for teams that value what the evidence says over what the demo shows.
§ Services
Five practices,
one collaborator.
Every engagement is delivered end-to-end by a single practitioner. No handoffs, no account layers — just the person building the thing.
- 01
Artificial Intelligence
Custom LLMs, fine-tuning, and AI assistants that turn unstructured knowledge into fast, reliable workflows.
- Retrieval-augmented systems
- LLM fine-tuning & evaluation
- Agentic tooling
- 02
Machine Learning
Deep neural networks and MLOps pipelines designed for production — reproducible, observable, and honest about their limits.
- TensorFlow / Keras models
- Kubeflow pipelines on GCP
- Model monitoring & drift
- 03
Applied Research
Graph neural networks, reinforcement learning and scheduling — bridging peer-reviewed research and shipped systems.
- Literature-to-prototype
- GNNs & deep RL
- Experiment design
- 04
Cloud Development
Full-stack systems on GCP, AWS and Azure — from data platforms and APIs to distributed schedulers.
- Data platforms & ETL
- Microservices & APIs
- CI/CD & infrastructure
- 05
Cybersecurity
Machine-learning-guided security research and hardening for containerised, distributed environments.
- Container security
- ML-assisted anomaly detection
- Threat modelling
§ Approach
A quiet preference for the measurable.
Phoenix Labs exists because interesting problems deserve careful work. It's a place to research, build, and share — a side project run with the same standards as any lab worth its name.
Principle 01
Science first
Decisions are grounded in measurement. If a model or system cannot be evaluated, it is not shipped.
Principle 02
Built to last
Systems are engineered for the team that inherits them — documented, testable, and boring where it matters.
Principle 03
Given back
Research, notes and tooling are contributed back to the community whenever the work allows it.
§ Selected work
Recent notebooks.
n = 3 of many
Research · Distributed Systems
KAIROS — GNN Scheduler
A framework that trains Graph Neural Networks from an ensemble of weaker heuristics, producing a scheduler that outperforms state-of-the-art baselines.
~ case study on request
Publication · Deep RL
DAG Workflow Scheduling
Peer-reviewed work on Actor–Critic reinforcement learning for scheduling directed acyclic workflows on shared infrastructure.
~ case study on request
§ Contact
Have a problem worth studying?
Consulting, research collaborations and open-source work all welcome. Short scopes preferred — long conversations included.
kleiton@phoenixlabs.com.brAlso on
- GitHub/ phoenix-labs
- LinkedIn/ phoenix-labs
- LocationRemote · Global
- Responsewithin 48h