Services
Two capabilities, shaped around your operating problem: mining and process intelligence, or AI connected to everyday business workflows.
Mining offerings · copper & silver flotation
Use existing historian and assay data. Scope the work around a named plant problem and advance only when the preceding layer is ready.
Assess historian and assay quality, sampling-window alignment, missing signals and plant-stop handling. Estimate demonstrated recovery headroom from comparable historical operating states.
Output: a data-readiness scorecard and plant-specific benchmark, followed by a recommendation on the next investment.
Review existing models, whoever built them, against a sixteen-item failure-mode checklist. Examine sensor health, validity, validation and operator adoption.
Output: a failure-mode scorecard and adoption review, including overrides, rejection causes and whether accepted recommendations were actioned.
Mining diagnostic editions: proposed 2–3 week fixed-fee engagements with historian access. Diagnostic offering remains in validation; scope is agreed separately from a build.
Soft-assay and stage models that make recovery visible during processing. A bounded deployment over the existing plant data layer.
Monitor failure modes on existing models, including sensor issues and operating changes. Retraining and promotion remain subject to review.
Constrained set-point recommendations presented as changes from current settings, with operator acceptance or rejection and reasons captured.
Advisory scope uses existing data and sensors. Recommendations do not write to plant control systems. The mining focus is copper and silver flotation.
A planned AI reasoning layer would use recorded rejection reasons and plant context to refine subsequent recommendations. This extension is designed, not a completed production result; it still routes through operator review.
01 / Data & AI
Make data reliable, develop useful models and connect the result to the people who need it.
Lakehouse and warehouse foundations, ingestion and streaming pipelines, semantic layers and data quality.
Forecasting, anomaly detection, recovery estimation and soft sensors. For process industries, operator-gated decision support using existing data.
ERP and CRM connections, analytics delivery and AI inside existing products.
02 / AI Engineering
Connect models, tools and business systems with a clear boundary around what AI can do.
Document intake, support triage and other bounded processes that require more than a chat interface.
Specialist agents coordinated across ERP, documents and collaboration tools.
Human approval gates, audit trails and explicit authority for consequential actions.
Where workflow AI starts
Invoice copies, statements and order enquiries handled through mailbox and ERP integrations, with clear exception and review paths.
Support triage reference ↗Extract request details, match products, evaluate pricing and assemble a quote. Keep commercial approvals with the account manager.
Quote Builder reference ↗Classify and extract documents, retain source references and flag conflicting values before a person approves the next action.
Document intake reference ↗Technical skills & engineering practices
We choose the stack around your data, integrations and operating requirements. These are technologies we work with, selected to fit the engagement.
Ingestion, streaming pipelines, warehouse and lakehouse foundations, semantic layers and data integration.
Python · PostgreSQL · BigQuery · Databricks · Snowflake
Forecasting, anomaly detection, computer vision, recovery estimation and soft sensors, with validation and monitoring.
Vertex AI · Azure ML · Amazon SageMaker
Retrieval-augmented generation, document workflows, specialist agents and orchestration connected to business tools.
OpenAI · Anthropic Claude · Google Gemini · Google ADK · LangChain · MCP
Web and mobile applications, backend services, ERP and CRM connectors, and human-review interfaces.
TypeScript · React · Flutter · Python · FastAPI · Node.js · PostgreSQL
Build and run applications in client cloud environments, connect existing systems and support the delivered workflow.
Google Cloud · AWS · Microsoft Azure
How the engineering holds together
Agree a baseline and evaluation set, check outputs against the workflow and measure the delivered system.
Human approval gates, auditable actions and clear boundaries around what an agent may read, propose or change.
Client-owned infrastructure, source code and runbooks, with monitoring and support responsibilities agreed.
We use tools including Cursor, GitHub Copilot, Claude Code and Gemini to support delivery, including test generation. The engineering team remains responsible for reviewing and validating what ships.
How we start
Start with the operational problem. Assess the alternatives before committing to implementation. Each stage is a separate commitment.
A proposed opportunity and readiness diagnostic for one workflow: baseline, data readiness, alternatives and a business case. The recommendation may be build, buy, partner, client-led or defer.
Diagnostic offering currently in validation.
A separately scoped build, from design through deployment and measurement. Discovery is separate from the 12-week build.
Monitor the delivered workflow, support operations and use evidence to decide what to improve or build next.