Institutional memory
Capture accepted lessons, specialist reasoning and operational context so knowledge remains available across team changes and handovers.
Eight connected layers
Reliable AI starts below the model.
Each layer has a defined owner, an agreed output and a testable handover to the next. Ownership and technology selections are confirmed for each engagement.
Intelligence · L5-L7 Data foundation · L0-L4
Your cloud or your premises
Hosting, data residency, identity and access, model infrastructure, integrations, monitoring and support are agreed with your IT team during discovery.
IntelligenceL5-L7
L7
AI applications
Evidence-grounded assistants, predictions and optimisation workflows.
L6
Knowledge graph
Connect rigs, equipment, events, lessons and contracts with source provenance.
L5
Context & policy
Machine-readable metric definitions, vocabulary, business rules and access rules.
built on
Data foundationL0-L4
L4
Curated data products
Gold-layer views with consistent definitions and documented data contracts.
L3
Cleaning & entity resolution
Validate, standardise and reconcile inconsistent records with human review.
L2
Raw data foundation
An immutable bronze layer, partitioned by source, rig and date.
L1
Ingestion
A pipeline per approved source, supporting batch and streaming patterns.
L0
Source inventory
Identify and profile authorised sources on the rig and on shore.
The AssureLens Cognitive Index
Turn operational knowledge into a reusable asset.
An intelligence layer that brings together your records, their meaning and the relationships between them.
PeopleExpertise, review responsibilities and decision owners.
ContractsTerms, obligations and supplier commitments.
ClientsClient requirements and authorised project context.
VendorsSupplier relationships, performance and risk evidence.
AssetsWells, rigs and equipment linked to operational history.
ProjectsProgrammes, activities, actions and lessons across wells.
FinanceAFEs, costs and commercial records connected to delivery.
SOPsProcedures, standards and approved ways of working.
Knowledge GraphIllustrative model - entities, relationships and access rules configured around your organisation.
01Prepare the evidence
02Connect the meaning
03Retrieve & reason
04Review & improve
A growing knowledge asset
New authorised records and reviewed connections extend the knowledge base, with quality checks keeping that growth useful.
Natural-language discovery
Find relevant evidence without knowing which report contains it. Results expose source links and distinguish missing evidence from supported findings.
Workflow automation
Agents use relationships to gather context, draft outputs and route agreed actions through configured review and approval steps.
Decision intelligence
Relate recurring incidents to assets, vendors, procedures and cost records to help specialists investigate causes and prioritise improvements.
Governance & traceability
Carry source references, access permissions, review status and change history so teams can inspect how a finding was formed.
An example question
“Which recurring equipment issues affected our offset wells, what actions were taken, and which procedures should we review before the next well?”
The index connects reports, equipment, actions and procedures to assemble relevant evidence for expert assessment. An illustrative query, not a live result.
Agent + SME collaboration
From project inputs to reviewed outputs.
External information supplements project records where access and usage rights are agreed. Expert review remains part of the workflow.
01
Your documents
Authorised records and project context.
02
AI agents
Extract, organise and connect the evidence.
03
Human review
SMEs validate relevance and challenge findings.
04
Structured outputs
Reviewed registers, plans, actions and reports.
Specialist agents
Risk & opportunity agents
Identify and categorise findings in unstructured records, retaining source citations.
Action tracking agents
Support action follow-up through status updates, reminders and assigned responsibility.
Lookahead agent
Prepare draft forward plans using risks, accepted lessons and operational context.
Report generation agent
Compile draft reports from agreed sources for the audience and review process.
The critical engineering task
Different records. One equipment identity.
Names, local identifiers, pressure units and maintenance codes can differ across systems. Joining them requires more than moving the data.
| Source | Equipment name | Local identifier | Unit | Maintenance code |
|---|---|---|---|---|
| System A | Mud Pump #2 | EQ-0042 | psi | MP-PM-90 |
| System B | Pump, Drilling Fluid 2 | 1180-B-02 | bar | 3.4.2 Quarterly |
| System C | No. 2 Slush Pump | SLP002 | kPa | PM-Q-MUD |
Resolution process
- 1Validate source fields and units.
- 2Apply matching rules to candidate records.
- 3Choose authoritative values through agreed survivorship rules.
- 4Route ambiguous matches to a human review queue.
- 5Publish the canonical identity with confidence and provenance.
Illustrative canonical record
Mud pump 02
- Canonical key
- EQ-MP-002
- Standard unit
- bar
- Match confidence
- 0.94 · example
- Source lineage
- Systems A, B, C
Illustrative records, not client data. 0.94 is not a universal acceptance threshold.
Technology choices
Fit the stack to the estate and the operating requirements.
Architecture choices to assess, not a mandatory bundle.
| Decision area | Options and purpose | What must be resolved |
|---|---|---|
| Hosting | Client-aligned cloud; Azure is the reference option. On-premises scope can be assessed. | Data residency, network boundaries, model hosting, infrastructure ownership, support. |
| Pipeline orchestration | Airflow or an equivalent for dependencies, scheduled work and recovery. | Operating skills, monitoring, retries, support, interaction with existing tooling. |
| Streaming | Kafka or a managed equivalent where event ingestion is needed. | Source rate, ordering, retention, buffering, replay and connectivity. |
| Transformation | dbt or an equivalent for tested, versioned data transformations. | Data contracts, lineage, quality checks and change management. |
| Data platform | Databricks or a client-aligned lake/lakehouse environment. | Storage, compute, scale, access, cost and existing investments. |
| Consumption | AssureLens workspaces, agreed APIs and tools such as Power BI. | Published datasets, metric definitions, refresh cycles, authorised audiences. |
| AI & retrieval | Document parsing, hybrid retrieval, knowledge graphs and specialist agents. | Model selection, source grounding, evaluation, human review, task limits. |
Security & control
Make the boundaries explicit.
Security requirements belong in the architecture and acceptance criteria from the start.
Access & isolation
Agree authentication, project roles, authorised sources and client data boundaries. Apply permissions to records and their derived outputs.
Data protection
Define encryption, retention, deletion, residency and permitted model processing with the client's IT and security teams.
Operational separation
Preserve the read-only rig interface. Keep decision support distinct from operational control and human sign-off.
Audit & change
Record source lineage, transformations, expert review and release changes. Define approval and escalation routes.
Delivery & acceptance
One accountable owner for every deliverable.
A foundation team owns L0-L4. AssureLens domain and AI specialists own L5-L7. Client owners approve definitions, access and release criteria.
0
Stage 0
Discover
The work
Inventory systems, owners, interfaces and representative data. Define the first use case.
How we know it's done
Source map, access plan, quality profile, scope and acceptance measures.
1
Stage 1
Ingest & retain
The work
Implement selected ingestion paths, raw storage, monitoring and recovery.
How we know it's done
Traceable source records, ingestion checks, demonstrated replay or recovery.
2
Stage 2
Clean & resolve
The work
Standardise identifiers, units and codes; implement quality rules and review queues.
How we know it's done
Reviewed match examples, quality measures, source-to-canonical mappings.
3
Stage 3
Curate & define
The work
Publish data products, metric definitions, vocabulary and access rules.
How we know it's done
Validated data contracts and reconciled measures signed off by owners.
4
Stage 4
Connect & evaluate AI
The work
Build graph relationships, retrieval and selected agents; test with records.
How we know it's done
Source-grounded outputs, specialist evaluations, permissions tests, documented limits.
5
Stage 5
Release & operate
The work
Train users, complete release checks, establish monitoring, support and change control.
How we know it's done
Approved release, named service owners, runbooks, escalation routes, support responsibilities.
The first technical session
Bring the people who know the data and the work.
A focused architecture session with your drilling sponsor, IT lead, data owners and AssureLens delivery team.
Before the session
Bring a source-system inventory, representative authorised records, existing metric definitions and security constraints.
During the session
Walk the layers, test entity-resolution assumptions, agree the read-only boundary and evaluate the stack.
Leave with
An initial ownership map, open technical questions and the scope of the discovery or first workflow.
Start a conversation
Which workflow should we improve first?
Bring a planning challenge, a set of operational records or a recurring review task.