Understand
Find the job worth fixing.
Watch the workflow, speak to the people doing it and identify the actual constraint. A slow process may need clearer ownership rather than another model call.
Ilhan Irem YuceFierce & Tame · Based in Malta · Built for real work
Fractional CAIO / Forward-deployed AI delivery
AI ambition is easy.
Make it operational.
The meeting went well. Everyone agreed that AI could help. On Monday, the same spreadsheet is still making the rounds. I work with the people, the process and the systems behind that gap—then build something the team can actually use.
Product judgment, hands-on implementation and commercial context. A fractional engagement, with scope and availability agreed up front.
01 / The operating model
Close to the problem. Responsible for the delivery.
Fractional CAIO is the leadership role. Forward-deployed is how I work: alongside the team, close enough to see where the process breaks and hands-on enough to build the first working version. The job continues through testing, adoption and handover.
Build
Connect the useful pieces.
Use the relevant data, APIs and existing tools to build a focused system. Define what it may read, change and send before granting access.
Prove
Test the uncomfortable cases.
Set a baseline. Check accuracy, failures, cost and human effort on representative work. A persuasive demo is a beginning, not an acceptance test.
Transfer
Make the team able to run it.
Document the system, its costs, permissions and exceptions. Give the operating owner a way to see failures and a clear route to recovery.
The test is Monday morning: does the work run better when nobody is presenting it?
02 / Work you can inspect
The portfolio is already running.
FreeMalta, MaltaInsider and Gambity are founder-built projects, not anonymous client logos. They show how I connect product decisions, publishing systems, research workflows and distribution. Open them. Read them. Ask how the work behind them runs.
2.17MGoogle Search impressions
16KGoogle Search clicks
282KGoogle AI feature impressions
75.7KBing Copilot & ChatGPT citations
FreeMalta results reported in my October 2026 portfolio/CV: the preceding 100-day reporting period, with no advertising spend. Impressions, clicks and citations are different measures; these are historical results, not a client forecast. Original evidence is presented in the portfolio.
Product & publishing
FreeMalta
A live information platform. The reading room demonstrates structured content, a consistent reading experience and a research-to-publication workflow.
Explore the live productEditorial operations
The News Beast
A news and digest product with repeatable publishing and distribution. The challenge is keeping sources, editorial decisions and delivery connected.
Explore the live productBusiness intelligence
MaltaInsider
A business-facing platform connecting research, commercial context and implementation conversations. Clear answers need a useful next step.
Explore the live productA new product category
Gambity
A prediction-market editorial product. Structured coverage, analyst voices and publishing workflows built around a distinct audience and subject.
Explore the live product03 / What we can build
Give AI a defined job.
Start where your team repeatedly loses time, context or quality. The system should earn its place in that workflow before the project grows into a platform.
Research & knowledge
From sources to an answer you can check.
Research pipelines, structured summaries and internal knowledge workflows. Keep source provenance, review rules and uncertainty visible, especially before a result reaches a customer.
Operations & handoffs
From a trigger to an owned next step.
Connect approved events to CRM updates, reporting or internal tasks. Build for duplicates, missing fields, retries and failures rather than just the successful path.
Product & content
From a brief to a publishable asset.
Design repeatable research, drafting, independent claim-checking and revision stages. Human editorial judgment remains part of the workflow where the subject or publication risk requires it.
AI adoption
From access to useful daily work.
Help teams shape repeatable use cases, review permissions and build operating guidance. Vendor selection, contractual controls and deployment requirements are assessed for the particular organisation.
Tools I work with: OpenAI Codex for hands-on product development; Researcher API workflows for research, verification and revision; OpenAI and Claude APIs; PHP; Cloudflare Workers, D1 and KV; Supabase; n8n and Make. The architecture follows the workflow, not the other way around.
An example / Research to publication
One request. A controlled delivery chain.
A research assistant that produces a fluent paragraph is useful. A repeatable production workflow also needs evidence, a quality threshold and a way to recover when a provider fails. This is the kind of operating detail I build around.
01 / Evidence
Research & source records
Define the brief, gather relevant sources and retain enough provenance to check important claims.
02 / Quality
Draft, verify, revise
Separate drafting from claim-checking. Route failures to revision or a person instead of quietly approving every output.
03 / Delivery
Publish & observe
Use structured output, status tracking and safe retries. Track costs and failures so a batch can resume without recreating completed work.
Control is a feature. An agent may prepare a draft or propose an action. Publishing, external messages, payments and sensitive data changes need explicit authority and approval boundaries suited to the organisation.
04 / The person in the workflow
I understand what sits around the software.
Twelve years in Malta. Commercial leadership, payments, KYC, AML, fraud and risk operations before building AI-native products. That background helps me ask the questions a demo tends to skip: who owns the result, what happens when it is wrong, and is the change worth maintaining?
My work connects product ownership with hands-on delivery. I use Codex to develop and iterate, APIs to connect capabilities, and research workflows to make content production reviewable. The point is a useful operating system for the work, with a person accountable for it.
Partner status: OpenAI Select Partner, as recorded in my current CV. Specific product access, security features, support terms and contracts depend on the selected service and agreement. Partnership is not a substitute for evaluating the implementation.
05 / A sensible first engagement
One workflow. A clear finish line.
Agree on the problem, access and success criteria before expanding the scope. A fractional cadence can bring leadership and implementation together without turning a small operating problem into an indefinite transformation programme.
Discovery
Show me the ordinary day.
Bring one process, the people involved, the current tools and a sample of the work. We agree what needs to improve and what should stay human.
Scope
Make the outcome testable.
Define inputs, outputs, permissions, baseline measures and acceptance criteria. Record dependencies and choose an implementation route.
Delivery
Build with the operating team.
Work through a first version, test on representative cases and adjust the workflow. The timetable depends on data, access and required reviews.
Review & handover
Leave a system, not a mystery.
Review results against the baseline. Document configuration, ownership and running costs, then decide whether to iterate, expand or stop.
Typically structured around one or two days per week, with on-site work in Malta and remote delivery agreed to suit the engagement. Monthly scope, fees, availability and notice terms are agreed in advance. No salary comparison or guaranteed ROI is used to justify the work.
06 / Before we talk
The practical questions.
What is a Fractional CAIO?
Part-time AI leadership tied to practical implementation. The engagement covers problem selection, workflow design, building, testing and handover. The scope and decision-making authority are agreed with your team rather than assumed from the title.
What does forward-deployed mean here?
Working close to the people who use the system: understanding the process, connecting the relevant tools and data, building the workflow and testing it in the real operating environment. It describes a delivery approach, not a claimed affiliation with another company.
Who is this service for?
Founders and operational teams with a specific workflow problem, an accountable business owner and access to the relevant systems. Experience in payments, risk and commercial operations is particularly relevant to regulated businesses, but the engagement is not limited to Malta or one industry.
What can a first engagement deliver?
A scoped workflow, baseline measures, a tested first implementation, operating documentation and a review of what to improve next. The delivery schedule depends on complexity, access, data readiness and required approvals; there is no fixed two-week promise.
Do you work with our existing tools?
That is the starting point. Review the current stack and supported connections before adding software. Vendor choice follows the use case, data requirements and operating constraints, not a requirement to buy a collection of partner subscriptions.
Can you work on-site?
Based in Malta, with a fractional cadence typically structured around one or two days a week plus agreed delivery arrangements. On-site and remote work, availability, scope and fees are agreed before the engagement starts.
Will AI replace our staff or run everything unattended?
Neither is a default objective. The aim is to improve a defined process. Set permissions, approval boundaries, monitoring and exception handling so people know what the system may do and when responsibility returns to them.
What happens when the engagement ends?
Handover is part of the design: business-controlled accounts where appropriate, documented workflows, relevant source and configuration, known costs, access arrangements and an identified operational owner. Third-party licences and support terms remain separate.
Ilhan Irem Yuce / Fierce & Tame
Bring me the workflow everyone works around.
The report someone rebuilds every Friday. The research that never becomes usable. The handoff that always needs another message. Tell me what happens today, what you would like to change and who depends on the result. We can start there.
Direct contact: ilhan@freemalta.com. No promise of a fit before understanding the work.