Espen
Symphony

Enter your credentials to view this document.

Invalid credentials. Try again.

For Brendan — Working Note — 22 May 2026

Brendan, written for you, not for DGE: What 15 June
Asks Us to Build

Below is the honest picture — what the maximalist version asks for, what I’ve verified is real, and what I’d commit to delivering. Aim: both of us walk into the DGE conversation knowing exactly what we’re putting on the table. Today is 22 May; 15 June is 24 days away.

01 — The Premise

What the Vision Doc Asks For on 15 June

Brendan’s framing, page 3 of the Blueprint — the highest-priority scope, production-grade, with executive surfaces shipping behind:

“The Abu Dhabi individual journey (Surfaces 1–4: assessment, IDP, learning activation, profile) is the highest-priority engineering scope and must be hardened to production quality by 15 June. Executive surfaces can ship behind it.” Product Vision & Build Blueprint, p. 3

Behind those four surfaces sit a backend engine, an integrations layer, an AI stack, a platform foundation, a compliance workstream, and a content/localisation effort. The sections below break each of these into discrete modules. The intent is not to litigate scope but to map it.

02 — Data Foundation

The Inputs

The pre-populated profile depends on data flowing in from several sources before any user sees a screen.

# Module What it is Why it’s product-scale
01 HR Systems Integration Pipeline pulling employee records from DGE’s HR systems into the profile store. Schema mapping, identity reconciliation, change-data-capture, data access agreements. Workday / SAP SuccessFactors integrations are 6-month projects in normal contexts.
02 LinkedIn Ingestion Pull employee LinkedIn profiles into the system. LinkedIn API access tier, OAuth per employee, ToS compliance, data refresh cadence.
03 CV / Resume Parsing Extract structured data from uploaded CVs and resumes. NLP/LLM pipeline, format variance (PDF / Word / scans), Arabic CV handling, confidence scoring.
04 Internal Workforce Data Ingestion Pull from HR-adjacent systems — LMS history, performance reviews, certifications. Per-source connector, transformation logic, conflict resolution between sources.
03 — User-Facing Surfaces

The Screens

Everything the employee or manager actually touches. Each surface is its own design+build workstream.

# Module What it is Why it’s product-scale
05 Welcome Experience Personalised onboarding, embedded video / podcast, walkthrough, Ask Symphony intro. Content production, video hosting, personalisation rules, bilingual narration.
06 Pre-populated Profile UI Review and edit interface for the AI-generated profile. Diff-vs-AI-suggestion UX, conflict resolution, audit trail of edits, bilingual fields.
07 Assessment Surface ~20-min self-assessment across Skills, Generic Attributes, Behavioural Competencies. Survey engine, progress tracking, autosave, branching logic, mobile-first responsive, bilingual instrument with RTL.
08 Evidence Upload UX Attach evidence (project refs, peer validation, certifications) for ratings ≥ 80%. File upload, peer-validation request flow, certification verification, per-employee evidence library.
09 Manager Review Queue Manager-facing dashboard of pending assessments with per-capability review. Queue management, SLA visibility, batch operations, role-context switching.
10 IDP Surface Individual Development Plan generation and presentation. Goal tracking, deadline management, manager-approval gate, progress visualisation.
11 My Academy Surface Personalised learning recommendations from integrated providers. Provider-agnostic course display, deep-linking, progress sync, bilingual UX.
04 — AI Surfaces

The Five Named AI Surfaces

From p. 7 of the Blueprint. Each one is a distinct ML product with its own data, prompts, evaluation, and guardrails.

# Module What it is Why it’s product-scale
12 AI Job Description Generator Drafts and refines role scope from prompts and HR data. Prompt engineering, in-environment LLM, JD-quality evaluation, refinement UX, bilingual output.
13 AI Role Mapping Maps employees to SFIA capabilities and proficiency targets. Embedding model on SFIA taxonomy, similarity scoring, confidence handling, RAG over framework.
14 AI Gap Ranking Impact-weighted composite ranking of gaps. Multi-input scoring function (criticality × contribution × strategic flag), tunable weights, explainability for managers.
15 AI Next-Best-Action Recommends specific learning and IDP actions per gap. Catalog ranking across three providers plus internal, gap-to-action mapping, 70/20/10 heuristic, freshness.
16 Ask Symphony Agent Conversational layer over employee’s profile, assessment, IDP. RAG retrieval index, conversation state, safety / PII guardrails, bilingual conversation.
05 — Backend Engines

The Brain

The scoring, framework, and assessment engines — plus the API contract that lets the experience layer talk to them.

# Module What it is Why it’s product-scale
17 Scoring Engine Weighted composite (50 / 30 / 20), per-capability proficiency, Score publication. Math accuracy, audit trail, versioning when weights change, replay capability.
18 Capability Framework Service Source of truth for SFIA v9 — capabilities, levels 1–7, descriptors. Framework versioning, capability metadata, taxonomy queries.
19 Assessment Engine Survey logic, response storage, state machine (Draft → Submitted → Approved → Returned). Event-sourced state, idempotency, retry semantics, time-in-state SLAs.
20 Engine API The seven endpoints on p. 11 — the contract between experience and engine. Auth, versioning, error formats, pagination, rate limiting, documentation, UI SDK.
21 Notification Engine Triggered emails — manager review, IDP reminders, SLA overdue, status changes. Template management, bilingual content, delivery tracking, opt-out, queueing.
06 — Learning Integrations

Three Provider Builds

Each provider integration is a multi-week project of its own — auth, catalog sync, recommendation calls, completion handling, error states.

# Module What it is Why it’s product-scale
22 Coursera Integration Course recommendation plus completion sync. OAuth, catalog ingestion, recommendation API, completion webhook, license / seat management.
23 Pluralsight Integration Same shape for Pluralsight. Same shape, different API and contract terms.
24 Udemy Integration Same shape for Udemy. Same shape, different API and contract terms.
07 — Platform Foundation

The Boring Expensive Stuff

The non-functional substrate that earns the “production-grade” label in the Vision doc. None of these are negotiable for a licensed deployment inside the DGE environment.

# Module What it is Why it’s product-scale
25 Identity & SSO Integration with DGE identity provider, SSO, MFA. SAML / OIDC negotiation, token handling, session lifecycle, MFA enforcement for elevated roles.
26 RBAC System Five-role permission model enforced at the engine API layer. Role definitions, scope-based access, n ≥ 5 suppression for executive aggregates, anti-tampering.
27 Audit Logging Every action logged, audit trail UI for HR, retention ≥ 2 years. Append-only store, query UI, retention / archival, AI-use logging.
28 i18n Framework + RTL System All UI strings, survey instrument, IDP content, emails, admin docs in EN and AR. Translation management, RTL CSS framework, Arabic typography (numerals, dates, names), translator workflow.
29 Sovereign Hosting Deployment inside the DGE environment, in-country UAE. Infrastructure design, deployment pipeline, hosting agreements, data residency enforcement.
30 99.9% SLA Architecture High availability, multi-AZ, auto-failover. Load balancing, DB replication, monitoring, runbooks, SLA reporting.
31 Disaster Recovery RPO ≤ 1 hour, RTO ≤ 4 hours, documented runbook, quarterly tests. Backup strategy, restoration tooling, regular DR drills.
32 Performance Engineering Hit the budgets — 1.5s p95 load, 3s submission, 5s IDP gen. Caching strategy, query optimisation, load testing rig, p95 monitoring.
33 Observability Stack Logging, metrics, traces, alerting. APM, log aggregation, SLA dashboard, on-call rotation.
34 Admin Console Platform config, integration management, role assignment, audit log access. Self-service for HR / Admin, no engineer-in-the-loop for daily ops.
08 — Compliance & Governance

The Audit Workstream

External audits with their own timelines, independent of the build. None of these can be compressed by adding engineers.

# Module What it is Why it’s product-scale
35 WCAG 2.1 AA Conformance Audit and remediation across both experiences. 2–4 week external audit, remediation cycles, conformance statement.
36 Third-party Pen Test Security assessment, remediation, sign-off. 4–8 week external engagement minimum.
37 UAE PDP Compliance UAE Federal Decree-Law No. 45 of 2021, Abu Dhabi data-classification standards. Privacy impact assessment, processing register, consent flows, breach notification process.
38 Encryption / Key Management At-rest, in-transit, customer-managed keys where required. KMS integration, key rotation policy, audit trail.
09 — Content Production

Often Forgotten — Also Product-Scale

Translation and content production are usually the schedule risks nobody books for. Worth surfacing now.

# Module What it is Why it’s product-scale
39 SFIA Translation (EN → AR) ~150 skills × 7 levels × 4 axes — thousands of strings. Government-grade Arabic, terminology consistency, SME review, sign-off.
40 Onboarding Video / Podcast Welcome experience media. Bilingual scripts, production, hosting, captions, accessibility.
41 Manager Training Content Materials to train managers on the rubric and review workflow. Rubric explainer, scoring guidance, evidence assessment, bilingual.
42 System Admin Documentation Reference materials for ongoing operation. Runbooks, troubleshooting, escalation paths, bilingual.
10 — The Picture

The Cluster View

Forty-two modules, grouped into nine team-owned product clusters. The right-hand column is a realistic team size for a credible 15 June production launch.

# Product Cluster Modules Realistic team size
01 Profile & Onboarding 01, 02, 03, 04, 05, 06 1 PM + 3–4 engineers + content / data
02 Assessment & Evidence 07, 08, 19 1 PM + 2–3 engineers
03 Manager Workflow 09, 17, 21 1 PM + 2 engineers
04 IDP & Learning 10, 11, 22, 23, 24 1 PM + 3 engineers + integrations
05 AI Surfaces 12, 13, 14, 15, 16 1 PM + 3 ML/AI engineers + ops
06 Backend Engine 17, 18, 19, 20 1 PM + 3–4 engineers
07 Platform Foundation 25–34 1 EM + 4–5 platform engineers
08 Compliance 35, 36, 37, 38 1 compliance lead + external auditors
09 Content / Localisation 39, 40, 41, 42 1 content lead + translators + producers
42 Modules
9 Clusters
~25 Engineers
25 Days
11 — After Your Confirmation

Why the MVP Is the Only Path

Three honest brackets when assessing whether a small team can ship a complex platform. The picture below cuts across all three, applied to the confirmed reality of what exists today.

  Bracket The status, confirmed
A What’s already built & reusable Per what you’ve shared: SkillsTX FUSION TX (the assessment engine), the Lovable prototype (visual design system, 52 components, exec dashboard chrome — not production-grade), and the Symphony visual language. Nothing else. No EMEA build team. No prior Symphony codebase. No reusable scoring or framework engine.
B What’s integration-bounded Schedules set by external parties that cannot be compressed by adding engineers: third-party pen test (4–8 weeks), WCAG audit (2–4 weeks), DGE HR integration window, SkillsTX API contract and sandbox access, Arabic translation workflow, sovereign hosting inside the DGE environment. Each runs on someone else’s calendar.
C What’s coordination-bounded Work that scales sub-linearly with team size: distributed delivery across ESPER + Kyndryl (HUMAIN-committed, not DGE) + DGE in three time zones; 30–40% throughput loss in the first month before working norms settle; cross-discipline handoffs; manager UAT cycle. Adding people would slow this down, not speed it up.

What This Adds Up To

This is why Section 13 is framed as the only buildable scope, not a negotiation starting position. The arithmetic doesn’t shift based on confidence — but I want this to land for both of us. Hold this scope and 15 June ships clean; widen it and the date moves.

12 — Pre-Flight Check

What I’ve Already Validated

Before I’d commit to the MVP scope, I needed to know what SkillsTX actually exposes and what of the Lovable build is reusable. Done. The summary below is from a live walk on 12 May (see Client Material/skillsTX-mobile-app-feasibility.md) and a code audit of richandfancy/remix-of-symphony-ai-insights. Three findings tightened the picture meaningfully.

  Finding What it changes for the MVP
A SkillsTX public API — concrete and scoped 40 endpoints across 7 clusters (account · jobs/roles · evidence · actions · relationships · badges · bulk SFTP). OAuth2 client-credentials, JWT bearer. The integration adapter (N1) is sized: ~5–7 endpoints to wrap, not a multi-week exploration.
B Magic-link is a working escape hatch GET /v1/account/{key}/link generates a deep-link that drops a user into the SkillsTX self-assessment UI authenticated as themselves. Module 07 doesn’t need to rebuild the rating screens. Native shell + embedded SkillsTX rating step (“Option A”) is feasible today.
C Arabic + RTL: SkillsTX already shipped it An undocumented internal API (observed live) exposes /api/contents/{lang} returning 13,504 fully-translated SFIA items in Arabic with inline RTL. /api/jobs/rolebasedsurveyjobs returns the 40 named roles. /api/languages confirms Arabic as a first-class option. Module 28 collapses from “translate the SFIA framework” to “translate ~100 UI strings.” Role taxonomy is fetchable, not hand-built.
D Lovable repo — partial reuse confirmed 52 Symphony visualisation components (HeroKpi, KpiTile, SkillHeatmap, GapList, RadarChart, ProgressRing, etc.) plus design-system tokens. React 18 + Tailwind 3.4 in the repo; our prototype is React 19 + Tailwind 4 — ~30–50% portable with light work. Cuts the Executive Overview (N2) build by roughly half. Design tokens and visual chrome carry over cleanly.

Two Questions Still Need SkillsTX to Confirm

Direct ask to Paul Collins / Prem Singh (named in the proposal team). Both can be answered in one email.

If both answers come back “yes” — we ship Option B (full native, no embedded chrome). If only the first — we ship Option A (native shell + embedded rating screens). Either way, the MVP in Section 13 holds. The Pre-Flight Check confirms we’re building on something real, not guessing.

13 — A Buildable MVP

What I’d Commit to Shipping on 15 June

Forty-two modules is the maximalist read. Below is what I’d commit to shipping by 15 June with the team that actually exists. The framing is deliberately narrow — under-promise, over-deliver. Hold this line and 15 June lands clean; widen it and the date moves.

The scope, in four lines
  • SkillsTX as the assessment base — no replacement, no parallel engine. We use what SkillsTX already does.
  • A mobile-first website that guides 150 DGE employees through SkillsTX’s SFIA v9 assessment flow.
  • SkillsTX’s standard report, rendered as a beautiful in-app screen — same data, dramatically better presentation.
  • A rudimentary workforce overview for executives — completion rates and headline gaps. No scenario modelling, no AI heat-maps.

What’s In

Six modules from the larger list plus two new ones — sized for the €12k UX engagement plus a focused integration effort on the engineering side.

# Module In what form for 15 June
07 Assessment Surface Native mobile-first shell wrapping the SkillsTX assessment flow via magic-link (Option A, verified feasible). Symphony-branded intro, progress, and return-to-app moments; SkillsTX’s own screens for the actual rating step. Upgrades to full native (Option B) if partner-tier API confirmed.
10 Report Screen Native render of the job-match payload from GET /v1/account/{key}/job/{jobType} — same data SkillsTX returns, redesigned mobile-first with clean radial chart, plain-language gap statements, and a single “next step” CTA. PDF download as a stretch.
09 Manager Review Queue Native list driven by GET /v1/account/{key}/relationships/employee + per-report job-match fetch. Acknowledge or comment per direct report. No per-capability rubric in v1.
25 Identity & SSO Pilot-level. DGE SSO if available before launch; magic-link invitations as fallback. No MFA in v1.
26 RBAC — Three Roles Employee · Manager · Executive. No n ≥ 5 suppression for the pilot — the exec view is anonymised by aggregation only.
28 i18n + RTL (UI only) ~100 UI strings translated to Arabic with RTL layout. Confirmed pre-flight: SkillsTX returns the 13,504-item Arabic SFIA content pack natively, so the framework itself is not in our translation scope.
N1 SkillsTX Integration Adapter (BFF) Cloudflare Workers + Hono BFF holding the SkillsTX client_secret. Wraps ~5–7 endpoints (account, job/list, job-match, evidence, relationships, magic-link), handles return-to-app deep-link. Engineering work, not UX — sits outside the €12k UX scope. Estimated effort: 5–7 days with AI-augmented build (Claude Code / Cursor agents on a tight TS codebase).
N2 Executive Overview Screen Single workforce view: completion %, headline top-N gaps, segmentation by role/department. Aggregation pattern: nightly batch via 150 × GET /v1/account/{key}/job/{jobType} calls cached in the BFF. ~30–50% of the chrome reusable from the Lovable repo (HeroKpi, KpiTile, SkillHeatmap, GapList).

What’s Out (for 15 June)

Real Blueprint capabilities — deferred to a Phase 2 conversation, not cancelled.

# Category Deferred because
01–04 Data ingestion Use a simple CSV upload of the 150 cohort. HR, LinkedIn, CV parsing all deferred — integration windows alone exceed our 25 days.
05, 06 Welcome video + AI profile setup Minimal intro screen; employee fills basic info inline if needed. No video production, no AI JD generator at pilot stage.
11, 22–24 Learning provider integrations Show SkillsTX’s native recommendations as links; no Coursera / Pluralsight / Udemy API integration in v1.
12–16 The five AI surfaces Use SkillsTX’s built-in scoring, mapping, and recommendation logic. Don’t build parallel AI surfaces yet.
17–20 Custom backend engine SkillsTX is the engine. We don’t build a parallel scoring service, capability framework, or assessment engine for the pilot.
21 Notification Engine Pilot-level email triggers only — assessment invite, completion confirm. No templated bilingual messaging system.
27, 29–34 Platform foundation (full posture) Pilot-level audit logging, hosting, SLA, DR, performance, observability, admin console. Formalised post-launch as the cohort scales.
35–38 Compliance & encryption Pilot PIA and statement-of-conformance. Formal WCAG audit and third-party pen test booked for post-launch (4–8 week windows).
39–42 Content production SkillsTX likely provides Arabic SFIA already. Short written guides for manager training; no video.

What We’d Need from You and DGE

14 — What June 15 Looks Like

Mobile Demo Strip

Seven screens that walk the full pilot flow — employee onboarding through to the executive view. Symphony brand palette (dark green · antique gold · teal), Lovable design system. Scroll horizontally.

Swipe through

7:42
symphony.dge.gov.ae
SymphonyAI
DGE GovDigital
Welcome to your
capability snapshot
A 20-minute, honest read of where you are and where you’re growing.
Sign in with DGE Use magic link instead
01Welcome · Sign-in DGE SSO or magic-link fallback. Minimal, branded, mobile-first.
7:43
symphony.dge.gov.ae
Good morning
Yasser.
We’ve already mapped your profile from DGE’s records — no forms to fill.
Your role
SVP, Head of AI
Technology Vertical · AI Engineering
47
skills
3
areas
~20
minutes
Start assessment Edit my role first
02Personal landing Profile pre-populated from HR. “We did the grunt work” — per Brendan’s May 19 framing.
7:51
symphony.dge.gov.ae · assessment
SECTION 1 OF 3 12 / 32
Professional Skills
Strategic Planning
How often do you set direction and influence strategic decisions in your domain?
Not yet — under guidance
Sometimes — emerging
Routinely — independent
Lead — sets direction
Save & continue Attach evidence (optional)
03Assessment in progress Native shell wrapping SkillsTX’s rating model (Option A). Bilingual EN/AR via SkillsTX content API.
8:05
symphony.dge.gov.ae · results
Your role fitness
76 %
Strong fit · 4 areas to grow
Skills
3.9
Generics
3.6
Behaviours
4.0
Top gap to close
Leading at scale
Now 2.4 → target 4.0
See your plan Download report
04Role fitness result Native render of SkillsTX’s job-match payload — same data, polished mobile presentation.
8:08
symphony.dge.gov.ae · plan
Your plan
3 priorities for the next 6 months
01 · Leading at scale
Executive Leadership in AI
Coursera · 6 weeks · ~4 hrs/week
02 · Stakeholder influence
Mentorship pairing
Match with COO this quarter
03 · Org design
Team Topologies
Reading · ~12 hours
Send to manager Add another action
05Development plan Top gaps paired with concrete next steps. Coursera / Pluralsight / Udemy as deep-links (no API integration in v1).
9:14
symphony.dge.gov.ae · manager
Manager view
5 awaiting your review
Ahmed Hassan
82%
SVP Data · 3 gaps · evidence attached
Layla Mahmoud
71%
Director, ML Platform · 4 gaps
Yasser Shaaban
76%
SVP AI · 3 gaps · 1 evidence pending
Khalid Al-Rashid
68%
Director Cloud · 5 gaps
+ 1 more
06Manager view Direct-reports queue. Per-employee acknowledge or comment. No per-capability rubric in v1.
9:31
symphony.dge.gov.ae · executive
Workforce at a glance
Technology Vertical
150 people · pilot cohort · updated daily
148
of 150 done
76%
avg fit
3.2
avg maturity
Top capability gaps
Leading at scale
2.1
Stakeholder Influence
2.4
Org design
2.5
AI governance
2.7
View full report Export CSV
07Executive overview Dark surface for command/strategy view. Aggregated daily, no real-time. ~30–50% chrome reusable from the Lovable repo.

These mockups visualise the eight modules in Section 13 in their composed form. Every screen renders against real data flowing through the SkillsTX adapter (N1). The dark executive view (07) borrows its visualisation chrome from the existing Lovable repo — not built from scratch.

15 — What I’m Putting on the Table

The Shape of the Conversation

Brendan — I’ll commit to shipping the eight modules in Section 13 by 15 June. The scope is tight on purpose. Under-promising is how I win, and how DGE gets a working pilot rather than a half-finished one. I’d rather over-deliver against this scope than miss a wider one.

Two things I need from you to hold the date: which of the eight modules are firm (what you don’t want to lose at the pilot stage), and how we staff the adapter work (N1). The adapter is engineering, not UX, and sits outside the €12k scope. Three options: expand the engagement, find a Kyndryl resource on the DGE side, or sequence it as a separate phase. Your call — but I’d want to know by end of this week to hold 15 June with any confidence.