A process engineer putting expert models and local LLMs to work on your site.
I'm George Bai, a mineral processing engineer turned data scientist. I have extensive experience in plant design and process engineering, and a PhD in modelling mineral processes.
Since then I have developed a flotation process twin for an operating mine, built vision models that read froth and ore from camera images, and built agent systems that a large energy company uses every day. Eleticle is where that becomes AI that runs on a plant's own hardware, so its data never leaves.
What can be built for your site?
Five service lines, covering software and hardware. Open one to see where it works on a plant. Everything runs on hardware the site owns or approves.
Discuss a project01Expert modelsProcess intelligence and optimisation
Process intelligence and optimisation
Predictive models, soft sensors, mass balance, simulation and plant performance diagnostics.
- GrindingA soft sensor for product size between lab samples.
- FlotationA process twin that tests a setpoint before the operator makes it.
- DewateringA model of cake moisture and solids recovery.
- LaboratoryA mass balance that reconciles assays with online readings.
02Local LLMsAI assistants and automation
AI assistants and automation
Engineering knowledge assistants, production reporting, troubleshooting and technical workflow automation.
- Control roomThe shift report, drafted from the historian and the log.
- Control roomA troubleshooting assistant that knows the manuals and past events.
- LaboratoryLab results checked against the online analysers.
03Expert models · visionComputer vision and monitoring
Computer vision and monitoring
Process observation, equipment condition monitoring and operational safety alerts.
- CrushingOre size and belt condition, read from the conveyor.
- FlotationFroth cameras that watch mobility and colour.
- TailingsSpill and restricted-zone alerts.
04HardwareAI hardware and infrastructure
AI hardware and infrastructure
Tailored AI servers, industrial edge computers, cameras and sensors, with on-site or client-approved deployment.
- Server roomAn AI server sized for the plant’s models and users.
- CrushingAn edge computer and camera at the conveyor.
- FlotationCameras and lighting over the cells.
- Control roomOperator stations that reach the models on the site network.
05ServicesConsulting, integration and support
Consulting, integration and support
Opportunity assessments, pilots, plant system integration, validation, commissioning, training and ongoing support.
- Control roomConnection to the control system and the historian.
- LaboratoryConnection to the lab system.
- Server roomCommissioning against a test plan.
- EverywhereTraining for the people who will run it.
Or take it as a product.
The intelligence layer for a plant, combining expert models with local language models. It watches cameras, checks lab against online readings, writes the shift report and answers engineering questions, all on the site's own servers.
What does a typical project look like?
Every project is different, whether it is consulting or Eleflow, and Eleticle shapes the work around what you need. The steps below are a typical path, not a fixed one.
Exploratory session
An open conversation about your site and where AI could help. No brief is needed.
Assessment and pilot
One opportunity is chosen, then proven as a working model or assistant on your historical data.
Integration
Connected to the systems you already run: historian, control system, lab and cameras.
Validation and commissioning
Checked against the plant, with a test plan, before anyone relies on it.
Training and support
Your people run it. The next problem starts only when this one works.
Who is behind Eleticle?
A career, drawn the way a plant is drawn. Three feeds at the top, one product at the bottom. Tap any unit to see what happened there.
- 2009 to 2014 · 2019 to 2022
Plant engineering
Extensive experience on the engineering side: coal preparation and flotation plant design with EPC and commissioning support, then commercialising patented mineral processing technology.
- Flowsheets, layouts, material and energy balances
- Projects in Australia, North America and South Africa
- A screening technology introduced to its region, published as a cover story in 2014
- 2017 to 2021 · The University of Queensland
Modelling a separation process
A PhD on centrifugal dewatering: first-principles models checked against experiments and pilot-scale tests, then carried into beta digital-twin software.
- Three industry-funded research projects
- Two journal papers on dewatering models
- 2017 to 2023
Teaching a camera to read a process
Image analysis and deep learning for process monitoring, tested on industrial equipment and not only in the lab.
- Froth mobility in an industrial flotation cell
- Gas dispersion from bubble images
- Online slurry solids measurement
- What the plant years left behind
Process judgement
Knowing what a recycle stream hides, why an instrument drifts, and where a model must not be trusted. It is the part of this work a data scientist cannot look up.
- Physics joined with data
Digital twins and machine learning
Hybrid models that keep the physics and let the data correct it, with optimisation on top: setpoints, scenarios and bottlenecks.
- Deep neural networks with plant data and simulation
- Particle swarm and genetic algorithm optimisation
- From research to site
Computer vision for monitoring
The research methods, applied on operations: ore-size estimation on a conveyor, and furnace and rolling-line monitoring that pairs image models with language models for reporting.
- 2022 to 2025 · global engineering consultancy
Twins for operating mines
Developed a zinc-lead flotation process twin that evaluates setpoint decisions, and verified it against the plant.
- Published at the Mill Operators Conference, 2024
- Value-chain optimisation for a gold mine
- Model-versus-plant verification and commissioning support
- 2025 to now
Language models put to work
Assistants that retrieve documents, call tools and work in teams, used daily for trading decisions in the energy sector, and language models deployed on an organisation’s own hardware.
- Now · Brisbane
Local AI for plants
Everything above, in one place: Eleflow, the product for mining and mineral processing teams, and custom AI built to run on a site’s own hardware.
- Process models that respect the physics
- Assistants that know the plant’s documents
- Cameras that watch what operators cannot
Where the experience comes from
Education
- PhD in EngineeringThe University of Queensland, 2021. Mineral processing, modelling and AI.
- Master of Engineering, Oil and GasThe University of Western Australia, 2017.
- Bachelor of Engineering, Mineral ProcessingChina University of Mining and Technology, 2009.
Work experience
- Founder, EleticleNow. Eleflow and custom AI for plants.
- Senior Data ScientistA large energy company, 2025 to now. Forecasting and LLM agent systems.
- Senior Data ScientistA global engineering consultancy, 2022 to 2025. A flotation process twin, value-chain optimisation and vision models.
- Senior Engineer and Project ManagerA mineral processing technology company, 2019 to 2022. Technology commercialisation, flowsheets and balances.
- Processing EngineerEngineering design firms, 2009 to 2014. Coal preparation plant design.
Selected publications
- Innovations in centrifugal dewatering of ultrafine coals and tailings.Australian Coal Preparation Conference, 2026.
- Practical considerations in the design of flotation process twins for set point decision automation at MMG Dugald River Mine.Mill Operators Conference, 2024.
- A convolutional neural network for classification of froth mobility in an industrial flotation cell.Mineral Processing and Extractive Metallurgy Review, 44(3), 2023.
- A model-based parametric study of centrifugal dewatering of mineral slurries.Minerals, 12(10), 1288, 2022.
- Measurement of solids concentration in aqueous slurries for monitoring the solids recovery in solid bowl centrifugation.Minerals Engineering, 170, 2021.
- Modelling solid-liquid separation and particle size classification in decanter centrifuges.Separation and Purification Technology, 263, 2021.
- Classification of gas dispersion states via deep learning based on images obtained from a bubble sampler.Chemical Engineering Journal Advances, 5, 2021.
- A touch of class.World Coal, 23(6), 2014. Fine-screen recovery of coarse coal slurry.