02 / PROJECTS

PORTFOLIO.

Atlas, Quid and Xeo.

00 / CONTEXT

Study
projects.

Many projects, products and creative concepts developed for employers are omitted from this portfolio for confidentiality reasons. This selection presents personal study projects: independent experiments with tools, frameworks and business strategies, developed to broaden the skill set alongside professional experience. The focus is applied research rather than launching new businesses.

Atlas. Aesthetic is a language. Say what matters the most. A mandrill portrait crossed by blue and red lines.

01 / STUDY PROJECT · E-COMMERCE

Atlas:Graphic clothing for animal lovers.

Atlas combines creative direction with e-commerce operations. The clothing’s visual language extends to the logo, design system and social channels; acquisition, onboarding and fulfilment complete the journey from interest to delivery.

DESIGN & BUILD / 01

From brand to store

Brand & designThe concept, logo and design system define the Atlas visual language across clothing, the store and social content.

E-commerce & operationsThe Shopify store brings together the catalogue, product pages, cart and checkout. Workflows cover onboarding, order management, external integrations and fulfilment automation.

Creative productionSourcing and coordination of freelance designers: briefs, revisions and asset adaptation through Fiverr and marketplaces such as Upwork.

Logo & design systemSocial identityShopifyOnboardingFulfillment

GTM / 02

Acquisition and conversion

Instagram drives discovery; Meta Ads re-engages interested users and supports the purchase decision. Content and audiences evolve with each funnel stage.

TOFU
Instagram
The social identity introduces the brand’s aesthetic and brings new visitors to the store.
MOFU
Meta Ads · retargeting
Campaigns bring people who have already engaged with the brand back to collections and product pages.
BOFU
Meta Ads · conversion
Retargeting focuses on purchase signals and order completion.

LIFECYCLE EMAIL / 03

Trigger-based email

Workflows designed around store events.

01 / SIGN-UPWelcome
Sign-up: brand and collections.
02 / INTERESTDiscovery
Product visit by an opted-in contact: collection content.
03 / CARTRecovery
Abandoned checkout: reminders after 4 and 24 hours if the order is still incomplete.
04 / POST-PURCHASERelationship
After delivery: garment care and a feedback request after 7 days.

Quid. Understand how you act. Make better choices. Every day. Mobile mockup of Quidz Pattern.

02 / STUDY PROJECT · DIGITAL PRODUCT

Quid.Coaching continues between sessions.

The concept responds to a market gap: the relationship with a coach is central, yet practice, discussion and observation between sessions often remain fragmented. Quid envisions a more continuous experience, combining individual coaching, a collective dimension and an everyday digital layer.

CREATIVE CONCEPT / LEARNING OBJECTIVE

Turn this insight into a product system: design the transitions between sessions, practice and community; explore how a mobile-first web app can make choices and behaviour visible over time; combine product design, development and AI in a new field. The project examines how to bring continuity to a service built on human relationships.

TECHNICAL SPECIFICATION / 01

Architecture and flow

Front-end React · Lovable: A mobile-first interface developed in Lovable with React: onboarding, HEXA journeys, Quidz, practices and dashboards. Components and interaction states translate the model into short, recurring actions.
Back-end and data Python · PostgreSQL: Python services handle application logic and content generation; PostgreSQL organises users, original responses, sessions, scores, profile versions and usage events. Data remains traceable from individual input to output.
Quidz generation Python · content pipeline: Python functions generate scenarios, choice options and practices from the model’s dimensions. Quidz responses produce observational signals about action, control, expression, adaptation and presence, separate from the foundational HEXA profile.
AI provider Claude Sonnet · OpenAI GPT-5: Python prepares structured context and calls the models via API to generate interpretations and natural-language content. Providers interpret previously calculated data and rules: they do not assign scores or modify the vector.

MODEL / FORMA VECTOR

A map of choices, not a label.

HEXA examines the factors shaping motivation, decisions and action from six perspectives. Responses are translated into normalised numerical coordinates: the Forma Vector combines them in a multidimensional, versioned structure, with weights, confidence levels and potential tensions between signals. Each interpretation can therefore be traced back to its underlying data. Quidz adds observations from everyday situations, kept separate from the HEXA core. AI translates these structures into understandable language; it does not calculate scores or rewrite the profile.

ResponsesScoringPostgreSQLForma VectorPython / AIOutput

SERVICE MODEL / 02

Three pillars, one journey

Each component serves a different moment in the journey. The relationship provides direction, the group offers support and the product makes practice possible between sessions.

01 / RELATIONSHIPCoach

Individual work gives context to choices, sets direction and interprets emerging patterns. Responsibility for the relationship remains with the person guiding the journey.

02 / PEER EXCHANGEThe Circle

A shared space for exchanging experiences and maintaining continuity beyond one-to-one sessions. Peer discussion makes practice less isolated and more sustainable.

03 / DAILY PRACTICEQuidfy

The web app brings the model to the phone: HEXA builds the Forma, Quidz and practices turn everyday situations into observable choices, while the programme keeps track of the journey.

PRODUCT DESIGN DECISION

Quidfy designs an interface between coaching sessions. Micro-interactions, interpretable data and guided content make practice observable without reducing a person to a score or automating the coach.

DISTRIBUTION / 03

Guerrilla marketing, social media and community nurturing as potential routes into the model.

Visit Quidfy

XEO banner: SEO, GEO and SERP ranking in one place, with an analytics dashboard mockup.

03 / STUDY PROJECT · SEARCH INTELLIGENCE

XEO.Digital visibility, read through evidence.

An SEO audit describes site structure; search data shows observed rankings; AEO/GEO analysis assesses how clearly content can be used in answers. XEO brings these perspectives together, separating measurements, interpretations and action priorities.

CREATIVE CONCEPT / LEARNING OBJECTIVE

Design an audit engine that turns scattered data into a clear basis for decisions: explore crawling, evidence models, provider integration, explainable scoring and reporting. The Python core remains independent of the interface shown in the mockup, allowing the same analysis to support different formats and channels.

TECHNICAL SPECIFICATION / 01

From website to report

Core Python CLI · main.py: A single entry point receives a public URL and device (Mobile or Desktop, with Mobile as the default). Orchestration coordinates collection, analysis, evaluation and reporting while keeping module responsibilities separate.
Collector requests · BeautifulSoup4: Robots.txt, sitemaps and internal links guide a representative sample of up to 50 public HTML pages. Each page is fetched once, normalised into SiteData and stored as JSON with a manifest of source and destination URLs.
Google CrUX · PageSpeed · Lighthouse: Core Web Vitals from real users are kept separate from laboratory tests. If CrUX has no data, the analysis continues with Lighthouse and states the measurement source.
DataForSEO Keywords Data · SERP LIVE: Candidate keywords are validated with market metrics; ten representative non-brand queries are submitted to Google SERP LIVE. The observed organic position, URL and device remain visible in the result.
Configurable LLM providers/llm.py: Receives only selected, cleaned textual evidence to interpret entities, topics, search intent and the content’s ability to answer questions. Its conclusions are labelled as inferences and kept distinct from technical observations and measured rankings.
Evaluation evaluation_engine.py: Evidence feeds into AuditResult: four scores, a technical eligibility gate, findings with severity and confidence levels, and prioritised recommendations. report_builder.py presents the result without recalculating scores.

URL + deviceCollectorSiteDataEngineAuditResultReport

ANALYSIS MODEL / 02

Four dimensions. One explainable index.

The XEO Score summarises the site’s observed readiness. Each dimension retains its own evidence and weight: the final number guides diagnosis rather than replicating a ranking algorithm.

40%Search Readiness

Crawl accessibility, metadata, architecture, content and structured data.

25%AIgen & LM Readiness

Entity clarity, evidence reliability and the content’s ability to answer questions.

15%Web Performance

Core Web Vitals and diagnostics, distinguishing real-user data from laboratory tests.

20%Keyword Position Performance

Observed organic rankings for ten non-brand queries, on the selected device and in the selected context.

SEPARATE GATE / TECHNICAL ELIGIBILITY

Issues affecting crawling or indexability are flagged before scoring: a critical status caps the XEO Score without hiding evidence from the other pillars.

METHODOLOGICAL RULE

Observed on the site, measured by a provider, derived from the market, inferred by the model or unavailable: every data point retains its source and confidence level. Recommended actions follow severity and impact, rather than a promise of rankings.