SELECTED WORK / VISUAL EVIDENCE

Evidence in view.

The operating models, analyses and prototypes behind the portfolio, made legible without publishing workplace records. A diagram should earn its wall space.

A PUBLIC VIEW OF THE WORK

Pictures should explain the decision.

These exhibits draw on working presentations and notebooks from 2022–26. Charts from the synthetic procurement, staff and student demonstrators are shown as original notebook or report outputs. The licence matrix and other diagrams are public reconstructions of my methods, because the source slides contain employer branding, colleague names, internal system details or restricted records.

Dates describe when the work was documented. A proposal, a prototype, a controlled test and a delivered analysis are labelled separately. The full decks, code and operational datasets remain privately held; I’m happy to discuss the methods and their limits.

01 / BA CAPABILITY

Building an internal consultancy.

The work began with a team vision and a practical resourcing approach, then matured into shared learning, measured activity and a defined service offer. The view below describes the approach without reproducing internal staffing or activity records.

People, demand, practice and a repeatable offer

10analysts I led and coached during the later capability period

50recorded uses of a BA resourcing request survey since 2022

75instances of project allocation or handover in that review

100+catalogued BA activities mapped to eight delivery gates by 2025

33 of 100 mapped activities

Detailed Design in the February 2024 activity snapshot; later catalogue scope expanded.

2022Vision and allocation

Team standards, mentoring and prioritisation criteria became explicit.

2022–24Community and evidence

Practice sharing, feedback and activity measures helped identify where support was needed.

2024–25Reusable methods

Persona, process and data requirements guidance made analysis easier to reuse.

2025–26Service catalogue

The offer was documented with outputs, effort cues and stage alignment.

Source and scope: 2022 team framework and resourcing deck; February 2024 milestone review; 2025 service catalogue. Resourcing and allocation counts reflect the author’s correction of 9 October 2026. The requests and allocation instances are different measures. Team size describes later leadership, and the catalogue count describes documented activities.

Data was built into the method: an August 2024 requirements draft made source, ownership, quality and validation explicit; a July 2025 strategy adapted them to four service domains. A December 2025 six-course learning pathway and February 2026 process approach followed. The proposed single portfolio view was a design goal, not a deployed data product. See the data assets ↗

Read the BA Capability case study ↗

02 / PERSONAS PLAYBOOK

Make complex services human.

A shared playbook connected roles, journeys and touchpoints so teams could ask who a change affected. The discovery and validation were collaborative. This view replaces internal persona sheets and stakeholder lists with the public logic of the work.

A journey, rather than a label

01 / DISCOVERRole and needWho is the person in this service context?

02 / MAPJourney and touchpointsWhere do access, information and support meet?

03 / VALIDATECheck with service ownersTest exceptions and wording with the people involved.

04 / MAINTAINReview as services changeOwnership and change control keep the playbook useful.

24 of 26 mapped and validated

Team status reported in February 2024; earlier phases used a different denominator.

Source and scope: 2023–24 persona discovery and launch presentations. A February 2024 portfolio update reported 24 of 26 personas mapped and validated at that snapshot; this was team work, with different scopes and denominators in earlier presentations. No internal persona cards or individual stakeholder details are reproduced.

PROCESS / OPERATING MODEL

Let project learning travel.

A later executive proposal connected the useful process model made for a project with an enterprise view of ownership, variants and automation readiness. It recommended a pilot before wider adoption. This diagram shows the proposed handoff, not an established enterprise platform.

From one project model to shared process knowledge

PROJECTMap the current workScope, handoffs, exceptions and evidence.

REVIEWCheck owner and versionPeer review, approval and change control.

CATALOGUEFind the reusable patternLink process, requirements and measures.

IMPROVEAssess candidatesPrioritise changes or automation with a sponsor.

Source and scope: BA process-management approach and later alignment executive summary. The integrated enterprise model was a recommendation with a pilot proposed, not a claim that every process was catalogued or automated.

Read the capability context ↗

03 / PROCUREMENT MODELS

What a model can show—and miss.

Process discovery gave the problem shape. A separate MSc demonstrator used fictional purchase-order records to investigate delays, approval stages, unusual patterns and carbon estimates. The gallery below shows unaltered chart outputs from synthetic-data notebooks. They show modelling work, not live financial results.

Feature importance chart from a synthetic purchase-order delay demonstration, with requisitioner experience and supplier lead time as the largest features — **Delay model: what influenced this synthetic fit**A feature-importance view can start a discussion about the model’s behaviour. It does not establish a causal explanation.

Synthetic test confusion matrix: 1 correctly identified no-delay, 34 false delay flags, 12 missed delays and 153 correctly flagged delays — **Failure modes are part of the evidence**One notebook run flagged most delayed cases but also 34 of 35 non-delayed cases. That limited synthetic test is unsuitable as evidence of operational readiness.

Synthetic urgency scoring feature-importance chart, dominated by a historical delays field — **Urgency needs scrutiny**Historical delays dominate this illustrative score. A real pilot would need to test whether that signal leaks future information or bakes past process failures into a new decision.

Synthetic scatter plot grouping sample requisitioners by experience and approval time — **Segments for a question, not a label for people**The clusters are fictional. In practice, staff-level grouping would need a legitimate purpose, fairness checks and human interpretation.

Precision, recall and F1 curves from a synthetic purchase-order delay classifier as its decision threshold changes — **Choose a threshold for the cost of each error**Precision, recall and F1 move differently as the cut-off changes. This synthetic run helps explain why a pilot needs an agreed human triage capacity, rather than a single impressive score.

Histogram of decision-function scores from a synthetic purchase-order isolation forest, with most cases clustered near positive scores — **Anomaly scores make a review queue**The isolation-forest output ranks unusual fictional orders. A score is a prompt to investigate, not an allegation of fraud or error.

The broader notebook experimented with eight applications. The public case study focuses on four chosen investigations because they formed the most coherent, explainable pilot scope. Both are demonstrators; neither represents a deployed scoring service.

Explore the procurement study ↗

03B / CARBON IMPACT PROXY

Put an indicative carbon signal beside spend.

In the MSc capstone, a synthetic purchase-order dataset was mapped from category and spend to an estimated kg CO₂e proxy. The source notebooks produced the category and department charts below. The colourful interpretation shows the calculation for an equal-spend example. This is a transparent factor mapping, not a validated prediction of supplier-specific emissions.

Illustrative £1,000 purchase by category; seven horizontal bars show assumed kg CO₂e from 800 for catering down to 50 for IT software — **Public illustration of the method.** Values are calculated from the illustrative factors in the capstone code, in kg CO₂e per £ spent. Equal £1,000 orders isolate the factor assumption; the bars are not real purchasing choices or measured footprints. The ±10–15% sensitivity discussed in the capstone is illustrative, not a statistical confidence interval.

Original capstone notebook output: total modelled carbon proxy by five categories across synthetic purchase orders — **Category totals in the synthetic notebook**This source chart sums assumed factor × synthetic spend. Large values reflect the fictional generator, category mix and modelling assumptions; they do not describe an institution’s footprint.

Original capstone stacked bars of synthetic carbon proxy by fictional department and category — **Where to ask the next question**The stacked view separates categories within fictional departments. It is useful for discussing data quality, factor provenance and procurement categories before any real-world carbon accounting.

The implementation uses example factors for catering, lab consumables, travel, hardware, facilities, software and other purchases. No lifecycle boundary, supplier-specific factor or externally verified benchmark is established by these charts. The underlying Python is privately maintained and I’m happy to discuss the calculation and its limitations.

Explore the capstone ↗

03C / STAFF SURVEY REHEARSAL

Plot the method; name the simulation.

The academic notebook prepared 2,000 synthetic staff records and built interactive views of survey themes, cohorts and risk. Its first static sentiment and topic charts are original outputs, but those particular labels were assigned randomly. They demonstrate rendering, not the state of any real workforce or a validated model.

Notebook chart of randomly simulated positive, neutral and negative sentiment categories across synthetic records — **A chart with a provenance test**Randomly simulated sentiment produces three bars. The plot is useful for checking the display workflow, not for interpreting staff feedback.

Notebook chart ranking ten randomly allocated theme labels in synthetic records — **Theme labels without inferred topics**These counts reflect random allocation, not natural-language topic modelling or actual organisational priorities.

The notebooks and reports describe richer predictive and pipeline ambitions. The figures shown here do not substantiate operational deployment, model accuracy or real employee risk. Read the staff and student research boundary ↗

04 / FROM ANALYSIS TO DECISION

Make a funding question answerable.

In a requisition and purchase-order change case, analysis connected validated pain points to a mapped current process, requirements and a proposed future state. A February 2026 presentation recorded that funding had been confirmed on 30 January 2026. This is evidence of the case’s progression, not a claim that the predictive models generated a saving.

The evidence that moved the conversation

DISCOVERY26 pain pointsValidated across business roles in the case material.

PROCESSMap the handoffsIdentify duplication, manual workarounds and control gaps.

DESIGN60+ requirementsMake a proposed service and delivery decisions testable.

DECISIONFunding confirmedRecorded on 30 January 2026; implementation remained a further stage.

Source and scope: February 2026 business-analysis presentation. Counts describe documented discovery and requirements, with underlying process maps withheld. The deck records a funding decision; it does not prove benefits realised, individual attribution or completion of the automation.

05 / MEETING AGENTS

A controlled test before a service.

The proof of concept connected meeting preparation, scheduling and follow-up with routing and integration roles. This public diagram replaces internal environment views and demo prompts, preserving the useful distinction between configured roles and tested behaviour.

Three task agents, two supporting roles

01Prepare

Briefing and links, with a fallback when sources are absent.

02Schedule

Candidate windows and conflict checks for human review.

03Follow up

Drafted summary and actions for the organiser to confirm.

04Orchestrate

Route the request to a suitable task agent.

05Integrate

Connect approved services within permission boundaries.

10/11Controlled checks passed

Reported on 27 October 2025. Further privacy and reliability gates remained.

Source and scope: August–November 2025 checkpoint decks. Five configured roles and ten successful checks in an eleven-case test set do not establish production deployment or live-service reliability. Internal endpoints, prompts, identities and permissions are omitted.

Read the meeting agents case study ↗

06 / RESTRICTED REAL DATA

Show the method; protect the records.

Three investigations used real, restricted operational exports. The screen below is a public reconstruction of the kinds of views an analyst used. Its bars have no relationship to the source values and should not be read as actual results.

ANALYST VIEW / PUBLIC RECONSTRUCTION

From an export to a decision

Choose a comparable measure, inspect coverage and missingness, then check the operational context before acting.

Example bars only · no workplace data

Real-data cases: Welcome Week survey analysis (three cycles, 2023/24–2025/26), software licence use (2025) and BA positive feedback (exploratory 2025). The first is descriptive analysis; the other two include exploratory model or rule ideas. No student response, user account, colleague score or commercial licence count is shown here.

STUDENT EXPERIENCEComparable themes

Harmonise questions and preserve missing answers.

Welcome Week ↗

LICENCE STEWARDSHIPRecency and context

Review under-use without auto-reclaiming access.

Licence analysis ↗

TEAM LEARNINGPositive feedback themes

Surface patterns without ranking people.

Feedback study ↗

DECISION RULEHuman judgement

Check source quality and consequences before a change.

My method ↗

06B / LICENCE STEWARDSHIP

Use the visual to ask, then verify.

The December 2025 usage notebook explored licence categories, last activity, content creation and candidate review. Its original charts encode restricted counts and user-level patterns. This new matrix retains the analytical idea with invented positions and no real values.

Fictional licence-use matrix contrasting activity recency and content creation, marked as illustrative and requiring human review — **Public reconstruction, fictional points.** Roles, leave, seasonal use and incomplete telemetry may explain apparent under-use. A candidate flag is neither a licence reclaimed nor money saved.

One source chart counted the month of each person’s last activity. That distribution is not a monthly active-user history, so its extrapolation is not a verified forecast. The real export, counts and Python code remain private. Explore the licence case ↗

07 / PROPOSED CAPABILITIES

Designs with a clear stage.

The campus wayfinding concept and Data Academy pathway were proposals. They show how I framed service design and capability uplift; the slides do not establish that a live campus platform or every planned class was delivered.

A wayfinding concept built in layers

DATASources and definitionsRooms, timetables, accessibility and availability.

TRUSTQuality and ownershipRefresh, consent, security and clear limitations.

SERVICEFind a usable spaceA coherent mobile and campus experience.

LEARNTest with usersDiscovery, pilot measures and iteration.

Source and scope: 22 May 2026 campus wayfinding business-case deck. This is a proposed approach, not a deployed interface or a claim of live occupancy prediction.

A planned learning pathway for the team

01AI and data literacy

Shared language and the questions to ask.

02Data science mindset

Understand the lifecycle and responsible use of generative AI.

03Python in practice

Work through an end-to-end example and discuss limits.

Source and scope: December 2025 capability-uplift plan. Six courses and roughly 29 hours of material were mapped into team sessions and self-study. Dates in the source were a schedule, not attendance or completion evidence.

DECEMBER 2025 / PLANNED PATHWAY

Six courses, three steps toward confidence.

A proposed route for the change team. The source planned roughly 29 hours of course material and seven booked three-hour team sessions; the difference called for independent study. This is curriculum design, not a completion chart.

01Data & AI literacyBig Data in the Age of AI
AI Fundamentals for Data Professionals

02Data science mindsetData Science Foundations
Generative AI for Data Analysis

03Python in practicePython Quick Start
Python for Data Science & ML

~29 hcourse material mapped

21 hteam sessions planned

Public redraw from the 8 December 2025 academy plan and capability-uplift presentation. Titles are abbreviated; no attendance or completion claim is implied.

See how these fit the BA Capability ↗

THE EVIDENCE BOUNDARY

Public charts and diagrams explain what was studied, built or proposed. Originals from the previous employer contain names, branding, internal data and working details, so they are not downloadable here. The original notebook charts above are real outputs from fictional records; the colourful factor illustration is a new public reconstruction of the calculation. I keep the Python code and source evidence privately and welcome questions about methodology, validation and what would be required for responsible use.

Return to the AI & Data portfolio ↗

EXPLAINABLE AI & DATA / READ THE METHOD

Read the evidence, understand the method.

A guide to the charts and case evidence on this page. Exhibit captions remain the authority for data origin and the claims each visual supports.

Follow the reasoning.

SourceRead the exhibit’s evidence label
RepresentInspect units, categories and scales
CompareDistinguish observed results from demonstrations
ReviewCheck limitations before taking action

Explore the building blocks.

AI Capability4

AI/Data Concepts2

Data Science Pipeline1

Evaluation & XAI3

Python Package2

Responsible AI1

Visualisation1

Number of concepts in this guide, not performance or usage measures.

Explore the explanation matrix · 14 concepts

Concept categoryFind a term or example

14 concepts shown

On small screens, scroll the table sideways. “Learning context” explains a related idea; it does not mean that method runs on this page. Deep learning is the standard term for “deep ML”.

Technical term	Technical description / definition	In plain English	How this page uses it
Visual encodingVisualisation	Mapping values or categories to marks, position, length, colour and labels.Primary reference ↗	Make the comparison visible, with a scale and a clear label.	Published analysis Read axis labels, denominators and captions before comparing exhibits.
Synthetic dataAI/Data Concepts	Generated records used to represent scenarios rather than directly report observed individuals or events.Primary reference ↗	A practice dataset. Useful for demonstration, not proof of workplace impact.	Demonstration Exhibits labelled synthetic or reconstructed illustrate methods; they are not disclosed workplace records.
Descriptive analyticsAI/Data Concepts	Summaries of observed data using counts, rates, distributions or comparisons.Primary reference ↗	Tell us what the records show; this alone does not predict next week.	Published analysis Counts and trend summaries describe the stated records and period, not future performance.
Residuals and prediction errorEvaluation & XAI	A residual compares an observed value with its fitted or predicted value. Error metrics summarise those discrepancies.Primary reference ↗	How far did the estimate miss, and where does it miss most?	Learning context Prediction exhibits can compare observed and estimated values; a smooth chart alone is not validation.
Confusion matrix, precision and recallEvaluation & XAI	A confusion matrix counts predicted versus actual classes. Precision measures positive-result reliability; recall measures positive-case coverage.Primary reference ↗	Which alarms were right, which were false, and which cases were missed?	Learning context Where a confusion chart is shown, inspect false alarms and missed cases as well as correct predictions.
Data science pipelineData Science Pipeline	An ordered, reproducible sequence for acquiring, validating, preparing, analysing and evaluating data.Primary reference ↗	A chain from a question and source to an answer someone can check.	Published analysis Case studies connect the problem, evidence, preparation, method, evaluation and decision.
Explanation versus causal proofEvaluation & XAI	A model explanation describes selected behaviour or contributions; it does not establish causal effects or full reliability.Primary reference ↗	A reason for this output is useful, but it is not proof that the model is right.	Responsible use A persuasive visual does not prove causation or transferability.
Data minimisationResponsible AI	Limit collected, disclosed and retained information to what is needed for the stated purpose.Primary reference ↗	Use the smallest appropriate amount of information.	Responsible use Restricted source records stay private; approved aggregates and labelled reconstructions support public learning.
pandasPython Package	A Python library for labelled tabular data manipulation and analysis.Primary reference ↗	A programmable workbench for cleaning and comparing tables.	Learning context A common Python option for tabular preparation; its inclusion is not a claim that this browser runs pandas.
MatplotlibPython Package	A Python plotting library for static, animated and interactive visualisations.Primary reference ↗	Turn numerical work into labelled figures that can be inspected and shared.	Learning context A common way to prepare Python figures. This page displays published assets rather than launching Python plotting sessions.
Machine learning (ML)AI Capability	Methods that estimate patterns or parameters from examples to support inference on other inputs.Primary reference ↗	Learn a pattern from examples, then check whether it helps with new ones.	Learning context ML learns from examples. Read the page-specific methods below to see whether any model runs here.
Deep learningAI Capability	Machine learning using neural networks with multiple representation-learning layers.Primary reference ↗	A layered model learns useful representations; more layers do not guarantee a better answer.	Learning context Deep learning is an AI capability, not a label for every chart. No deep model is implied by this explanation panel.
Natural language processing (NLP)AI Capability	Computational methods for representing, analysing or generating human language.Primary reference ↗	Help a computer work with words, while remembering that context matters.	Learning context Text search, translation, thematic classification and generation are different language tasks; their implementation is identified separately.
Sentiment analysisAI Capability	A language-analysis task that estimates an expressed attitude, often as polarity or sentiment categories.Primary reference ↗	Estimate the tone of words—not the private feelings or personality of their author.	Learning context Sentiment estimates expressed tone. It must not be confused with topic classification or a diagnosis of a person.