LIACC institutional website — project memoir

This is the English companion to the main Portuguese report. It summarises the project for international reviewers. The full version, with all sections and rationale, is available at /report/ (PT-PT).

• Course: Scrum & Roadmap Labs for Digital Transformation (UMaia, 2025/2026).
• Addressed to: Prof. João Almeida (regente), Prof. Carlos Alves, Prof. Samuel Anjos.
• Student: Gabriel Lima.
• Original group: eight members (see §3 · Method).
• Client: LIACC — suggested by the instructor in class; accepted by the group.

Classical Scrum for planning (Product Vision, prioritised backlog, 8 epics, 6 sprints, Definition of Done, RACI) recorded in Notion; continuous Kanban for execution. Solo execution is acknowledged openly: the group agenda could not be synchronised, so the author delivered the work alone, preserving the Notion hub as the original group planning artefact and documenting how each Scrum obligation was met.

The author acted as a technical consultancy: the choice of stack is a supplier responsibility, not a client one. The content and interface were first modelled in a dependency-free Python static-site generator (a Next.js track was explored first, then dropped) used as a high-fidelity prototype and single source of truth. On that basis, the final delivery was migrated and published on WordPress, as required by the brief: the institutional site is live at liacc.gablima.com with a custom liacc theme, Custom Post Types, taxonomies, and ACF. Content is editable independently in Gutenberg and wp-admin. Execution and the final audit are detailed in §9 of the Portuguese report.

• Institutional site live on WordPress with a custom liacc theme at liacc.gablima.com.
• Native content model — Custom Post Types (person, project, news_item, education, unit), shared taxonomies (research_area, thematic_line), and ACF fields, with Person ↔ Publication ↔ Project ↔ Area ↔ Thematic ↔ Unit cross-links.
• About 123 published routes (hierarchical pages, CPT singles, taxonomy archives, blog, landings), IA-aligned URLs, and 301 redirects from the old site.
• Blog of 20 original posts on contemporary AI topics.
• Shop with products and an E-Goi "reserve interest" flow — there is no checkout; reservation is by interest sign-up, with an explicit "prototype" disclosure.
• People with real email and "About me" bios imported from the official LIACC profiles into WordPress; self-contained pages with no outbound links to the old site.
• Authenticated user area with saved topics and preferences, wired to the E-Goi funnel.
• E-Goi integration via a server-side proxy (API key in a wp_option, never client-side), single opt-in, and an RGPD proof-of-consent log.
• Tokenised design system with dark mode and font-size accessibility, carried into the WordPress theme.
• SEO via Yoast (focus keyphrases, meta descriptions, schema/JSON-LD, EN+PT XML sitemaps, canonical, hreflang); audit with practices implemented and prioritised recommendations.
• Bilingual PT-PT / EN with Polylang real PT pages, language toggle, and hreflang.
• Hardened security headers (HSTS, nonce-based CSP, X-Frame-Options, and others), blocked user enumeration, rotated admin credentials.

The individual component was delivered as a working MVP integrated into the live site, not as a proposal: LIACC Companion "Lúmen", an assistant represented by a lightbulb mascot (animated SVG launcher, navy-and-green brand identity), bilingual (PT/EN) and CSP-safe. Its thesis is that the primary purpose of an academic site is to build and signal scientific reputational capital, which converts into three returns: funding, talent, and partnerships.

Lúmen offers three modes: Descobrir ("Recommended for you" from the account's saved topics via WP_Query, no LLM); Orienta-me (a persona navigator — student/researcher/industry/international — routing to the right pages); and Conversar (a grounded chat that searches the site's own content via WordPress search, with a free LLM — Groq, Llama-3.3-70B — answering only from those pages, with citations and an explicit "I don't have that" fallback). Memory is the logged-in account; only public content reaches the LLM; the key is server-side; the proxy is provider-agnostic (it can switch to Claude later).

Honest scope note: the deck's earlier "RAG over all LIACC articles" was deliberately scoped down to site-content retrieval for feasibility, keeping answers grounded and cite-able. An SEO blog drives acquisition and registration-as-utility drives retention, feeding the same E-Goi funnel documented in the marketing deliverables. Target metrics (≥12% sign-up rate, ≥40% of sessions touching the Companion, ≥3,000 organic visits/month, ≥10 qualified contacts/month) are goals to be measured once analytics are in place, not measured results. The full treatment is in §10 of the Portuguese report.

The WordPress (40%) and E-Goi (10%) rubric components are implemented and live; all verticals were verified in production. Already resolved since the first delivery: first-party cookieless analytics installed (RGPD-friendly, with a wp-admin dashboard), database backups configured, credentials rotated, two Site Health criticals fixed (PHP session blocking the REST loopback), and the people directory aligned to the official public page. The conversational assistant was hardened for factual faithfulness (multilingual semantic retrieval + a verified-facts block + a 2nd-pass faithfulness verifier). Honest remaining items: a small SEO/readability "tail" (~36 items amber, not red — real bios and short product pages); Brotli/AVIF pending (host configuration); menu and footer edited in the theme rather than wp-admin; and a Mod_Security exclusion for /wp-json/liacc/v1/ to request from the host. Detailed in §9 of the Portuguese report.

The full narrative, with every design token, architectural decision, and rubric mapping, is available at /report/ (Portuguese).