Selected work in strategy, AI-assisted decision-making and software delivery, export compliance and operational coordination.
Export compliance for global semiconductor testingSpecialist export-compliance support for a semiconductor testing business operating internationally.
A global semiconductor testing company engaged SemiSphere for export-compliance support. The engagement connects two areas of our expertise: the semiconductor industry and technology compliance.
Keep international business moving with informed decisions
For leaders in semiconductor testing, export compliance is part of managing international customer relationships and technical operations. Commercial decisions need to account for the technology involved, the work being requested and the circumstances in which it will take place.
The business need is practical: understand which opportunities can proceed, recognise where specialist review is needed and give the people making decisions a sound basis for doing so.
Connect industry knowledge with compliance expertise
This engagement illustrates why sector knowledge matters in technology compliance. Understanding the technical and commercial context helps specialists ask relevant questions and relate compliance considerations to the way a semiconductor business operates.
SemiSphere combines semiconductor advisory with technology-compliance expertise to support those discussions. The emphasis is on informed decisions that account for both commercial priorities and compliance responsibilities.
Evidence-led decision support with multiple AI perspectivesAssess competing arguments, challenge assumptions and trace recommendations back to supporting information.
Assess competing arguments, challenge assumptions and trace recommendations back to supporting information.
Business decisions involve incomplete information, competing priorities and knowledge spread across documents and systems. AESIR brings together structured AI deliberation and a retrieval-augmented generation (RAG) pipeline, with an architecture designed to connect the discussion to relevant evidence.
Designed and built by Martijn Dekkers, AESIR coordinates a council of AI models from different providers to examine a problem, challenge contributions and produce a consolidated assessment for human review.
Make organisational knowledge part of the analysis
Decision-makers need recommendations that account for their circumstances: the information available, the constraints they face and the evidence behind a proposal.
AESIR’s RAG pipeline is designed to retrieve relevant material and bring it into the council’s discussion. Document ingestion, indexing and retrieval provide the foundation for finding information that bears on the question. Source references connect retrieved material to its origin, allowing reviewers to examine the supporting evidence.
The intended benefit is analysis informed by the organisation’s knowledge and relevant research, with less manual effort spent gathering and repeatedly supplying context.
Research the questions that emerge
A discussion can expose missing information or assumptions that need checking. AESIR’s research design allows individual council members or dedicated research agents to request evidence as those questions arise.
This supports investigation from different perspectives and reduces dependence on one model’s selection of research. The approach distinguishes retrieved facts from inference and assumptions, helping reviewers understand which conclusions have supporting evidence and which require further investigation.
Challenge the options before committing
AESIR structures the discussion around the problem, assigns roles and coordinates contributions across debate rounds. Participating models examine the question and respond to developing arguments before a synthesis stage brings the assessment together.
For a technology investment, this approach can support examination of delivery effort, operating cost, architectural constraints and risk within one coordinated review. The decision-maker can assess the recommendation alongside the arguments and evidence that informed it.
Keep the work reviewable and controlled
The core deliberation pipeline has been implemented and tested, with an audit trail recording contributions, decisions and resource use. Budget limits are enforced within the orchestration, and the system is designed to halt safely while preserving partial results.
The RAG and research architecture extends that reviewability to the relationship between claims and their sources. Agreement between models remains one signal within the assessment; reviewers still need to judge the quality of the evidence.
The business value
AESIR is designed to make complex analysis easier to organise and scrutinise. It combines access to relevant knowledge with structured challenge, giving leaders a clearer basis for comparing options and identifying unresolved questions before committing resources.
Control and continuity in AI-assisted software deliveryCoordinate specialist AI agents, keep work moving across sessions and require evidence before accepting completion.
Coordinate specialist AI agents, keep work moving across sessions and require evidence before accepting completion.
AI-assisted development can produce code quickly, but delivering a working system also requires planning, architectural decisions, research and verification. As work grows, someone must manage dependencies, maintain context and establish whether the result meets the requirement.
OCHI, OpenCode Heavy Industries, is a development harness built on OpenCode. It coordinates specialist agents within a managed workflow, with the aim of reducing manual coordination while keeping progress and acceptance reviewable.
Give each part of the work a clear owner
OCHI separates planning, architecture, documentation research, codebase investigation and implementation. A coordinator manages dispatch and workflow progression, while specialist agents focus on their assigned responsibilities.
This structure helps connect the requested outcome to the work needed to deliver it. Planning identifies tasks and dependencies; research establishes relevant technical information; architectural decisions guide implementation.
For engineering leaders, the intended benefit is clearer accountability within an AI-assisted process: knowing which role is responsible for a decision, what work depends on it and what remains unresolved.
Run independent work together
Waiting for unrelated tasks to finish in sequence can slow delivery. Starting dependent tasks too early can create rework.
OCHI uses a dependency graph to identify work that is ready to proceed. Independent tasks can run in parallel, while dependent work remains blocked until its prerequisites are satisfied. Controls around overlapping write scopes help coordinate agents working in the same codebase.
The aim is to use parallel execution where it is useful while preserving the ordering and coordination the project requires.
Preserve continuity through interruptions
Long-running development work needs to survive beyond a single conversation or process. OCHI supports persistent role conversations and recovery, allowing work to retain continuity across sessions and restarts.
Operators can consult specialist roles directly, including planning and architecture, while retaining those conversations within the wider workflow. This supports focused intervention without requiring the entire task to be restarted.
Require evidence before progressing
OCHI’s workflow engine enforces progression through defined stages and evidence requirements. An agent’s statement that work is finished does not, by itself, satisfy those requirements.
This gives teams a mechanism for connecting completion to recorded verification and acceptance. It also makes outstanding requirements visible before subsequent work proceeds.
Demonstrated capability
The recorded v0.1 acceptance run passed seven black-box scenarios comprising 188 assertions. These covered persistent role conversations, planner consultation, parallel task execution, evidence-gated progression, end-to-end acceptance and restart recovery.
These results demonstrate the tested workflow capabilities. The business objective is to make AI-assisted delivery easier to supervise: less manual coordination between agents, clearer dependencies and a reviewable basis for accepting completed work.
Making export compliance manageable across products and marketsReduce repeated investigation, make export decisions easier to review and retain the evidence behind them.
Reduce repeated investigation, make export decisions easier to review and retain the evidence behind them.
For businesses selling internationally, export compliance brings together technical product information, regulatory requirements and the circumstances of a proposed transaction. When that information sits across spreadsheets, datasheets and email, teams spend time repeating checks and reconstructing the reasons behind earlier decisions.
ExportSentry was designed and built by Martijn Dekkers to bring classification, export-decision workflows and supporting evidence into a shared workspace. Available today, it serves multiple industries, including semiconductors, electronics, industrial equipment and engineering.
SemiSphere Group acts as a sales representative for ExportSentry.
Connect technical information with business decisions
Commercial teams need to understand whether an opportunity can proceed, what requires further review and which information is missing. Engineering and compliance teams need a consistent way to examine the product characteristics relevant to that assessment.
ExportSentry connects structured product information with explicit regulatory parameters and guided classification workflows. Teams can examine the basis for an assessment: the information supplied, the criteria considered and the rationale recorded.
Semiconductors are one application. The same approach supports businesses working with components, finished products and equipment across international supply chains.
Carry the assessment through to the transaction
ExportSentry links component classification to product records and subsequent export assessments. Teams bring together the relevant destination, parties, intended use and supporting documentation, with review steps for unresolved questions.
Decision-makers gain a clearer route from technical assessment to an accountable business decision. Engineering, compliance and commercial teams work from connected records, keeping the supporting information available as responsibility passes between them.
Keep the evidence with the decision
Responding to a customer query, internal review or audit should not require searching individual inboxes to reconstruct past work.
ExportSentry retains classification rationale, supporting documents and decision records together. Evidence packs and exportable reports communicate the assessment while preserving the underlying material for review.
When responsibilities change, another person can examine the recorded reasoning, understand the decision and identify what needs to be reassessed.
Automate repeatable work while retaining specialist judgement
Guided decision trees and explicit rules structure the classification process. Document management and evidence generation reduce the administrative work surrounding each assessment.
Human review remains part of the workflow, particularly where information is incomplete or interpretation requires specialist judgement. Teams can identify unresolved questions and retain the reasoning behind their decisions.
The business value
ExportSentry gives businesses a consistent, traceable process for assessing international opportunities and managing the compliance work required to pursue them. Connected records and repeatable workflows help teams spend less time reconstructing information and more time resolving the decisions that need their expertise.
Semiconductor acquisition due diligenceMarket insight and specialist access for an investment firm.
The requirement
An investment firm conducting commercial due diligence on an acquisition needed to understand the target company’s market and its specialist area within the semiconductor industry.
Our contribution
SemiSphere provided market insight and connected the client with a CEO and technologist working in that niche.
Client feedback
The client said this support helped them understand aspects of the business that would otherwise have been difficult or impossible to assess.
Clear oversight of airline crew testingLess time chasing updates. Clearer accountability. Better evidence of what happened and when.
Less time chasing updates. Clearer accountability. Better evidence of what happened and when.
Coordinating airline crew drug and alcohol testing involves several organisations, airport locations and time-sensitive handovers. CDATS, the Crew Drug and Alcohol Testing System, is designed to help operational teams manage that process, giving them a shared view of each case from initial assignment through notification, arrival and collection status.
The operational challenge
The workflow depended on paper packets, telephone calls and updates spread across separate systems. Coordinators needed to confirm that selected crew members were on the expected flight, assign someone to notify them and establish whether they had reached the collection location. Collectors had limited visibility of who was coming next.
Reconstructing events afterwards presented another problem. Notification evidence, timestamps, signed paperwork and billing information needed to be brought together to establish what had happened. Inconsistent records made exceptions harder to investigate and created additional administrative work.
A shared view of the work
The CDATS pilot scope connects these handovers through a digital case record. Coordinators can follow assignments and notification outcomes, while collection staff can see expected arrivals and record progress.
The design captures timestamps and supporting evidence as work takes place, including notification confirmation, arrival, exceptions and signed-form status. This gives teams a basis for identifying stalled cases and following up with the person responsible.
For operational leaders, the intended benefit is clearer oversight: understanding where work is accumulating, where intervention is needed and how incoming demand relates to collection capacity.
Evidence captured during delivery
CDATS is designed to retain the operational evidence needed for review, audit and billing support. QR-linked records, notification evidence and elapsed-time tracking connect individual actions to the relevant case.
Capturing this information during the workflow is intended to reduce the effort required to reconcile paperwork afterwards and make it easier to investigate delays or disputed events.
Working with established systems
The proposed approach preserves the organisation’s existing testing infrastructure. CDATS coordinates the operational workflow alongside WorkforceQA, FormFox and TRAQ, with laboratory results and medical determinations remaining in their established systems.
The initial pilot was scoped around operations at Dallas Fort Worth airport, providing a defined setting in which to assess the workflow before wider adoption.
The business value
The use case targets more predictable coordination, better visibility of incoming work and less time spent chasing or reconstructing information. It also creates a basis for assessing notification-to-arrival times, collection flow and exceptions, helping management decide where further improvements are needed.