RWT Work

Design research / October 2026 / both.

One problem.
both. sides.

How a recruitment idea travelled through sixteen visual directions, a two-voice prototype, several useful failures and towards something much more conversational.

This is the thinking, not the product launch.

Follow the design trail Back to experiments
Design exploration16 deliberately different initial visual studies
Usability feedbackRole-specific workspaces, fewer distractions and clearer actions
Current directionConversations, context and accountable next steps at the centre

01 / The problem

Recruitment is two people trying to make a good decision. The software often forgets one of them.

My starting point wasn't a new way to scrape CVs or calculate a mysterious percentage of human suitability. It was frustration with unanswered applications, opaque decisions, repetitive forms and recruiters trying to act on incomplete information. Both sides are expected to do the right thing without always being given the context to do it.

The working name arrived from a simple product test: does this make the experience better for both? Not identical screens, identical responsibilities or forced symmetry. Better information, better communication and a clearer account of what has actually happened.

A relationship isn't improved by making both people fill in the same dashboard.
01 / A useful beginningStart with very little.

A private candidate can explore relevant role types without a mandatory CV or fictional match confidence.

02 / Clear interestAsk rather than assume.

A recruiter who sees a promising but incomplete profile should be able to ask one relevant human question.

03 / Evidence, not theatreReceived isn't reviewed.

Application events, recruiter outreach and actual hiring decisions must be distinct and traceable.

04 / Honest uncertaintyUnknown is allowed.

Missing salary, a parsing gap or an absent CV must not silently turn into a negative verdict about someone.

02 / Visual divergence

Sixteen directions. The same two moments.

Before settling on a design system, I explored sixteen quite different visual compositions. Each had to handle the same pair of situations: a candidate considering a role and a recruiter recognising potential. This made the comparison more useful than sixteen attractive but unrelated landing pages.

Some leaned on enormous type, some on wayfinding, some on editorial sequencing, some on joined perspectives. The goal was to escape the usual job-board chrome without making the actual work harder.

Study 02 / Early favourite

Parallel Voices

Two distinct perspectives with a shared visual rhythm. It had energy, but equal presence everywhere later made the role switch confusing.

Study 06 / A different question

Dialogue

Conversation expressed through composition rather than chat bubbles. Useful as a principle, but typography alone couldn't carry complex workflows.

Study 13 / Shared identity

The Weave

The sides interleave instead of living in silos. A compelling idea, provided the joined visual language doesn't obscure each person's actual permissions.

Study 14 / Graphic restraint

Stencil / Space

Immense type and negative space. Striking as an opening statement, less convincing as the organising principle for an everyday recruitment conversation.

These panels are typographic reconstructions of documented concepts, not historical screenshots or a functioning UI. The initial Parallel Voices palette included coral for candidates, indigo for recruiters and a connecting lime accent.

The complete first pass

  1. 01Type as PlaceLarge text as navigation
  2. 02Parallel VoicesTwo connected perspectives
  3. 03IndexProfessional correspondence
  4. 04SignalWayfinding with minimal noise
  1. 05Stage CutAsymmetric progression
  2. 06DialogueConversation as composition
  3. 07CirculationPeople and roles in motion
  4. 08Night SignalBold after-dark editorial
  1. 09Open LedgerTransparent records and evidence
  2. 10Window + MaskLayered attention
  3. 11Vertical RhythmMobile-first editorial beats
  4. 12Restrained EditorialQuiet clarity
  1. 13The WeaveTwo flows, one system
  2. 14Stencil / SpaceTypography and negative space
  3. 15The Personal StageOne focused task
  4. 16Human DirectnessWarmth without ornament

The first decision was to explore Parallel Voices in greater depth, not to approve a complete identity system. Dialogue, The Weave and Stencil / Space stayed useful comparative directions.

03 / First implementation

Then I put it on a phone. The design got less clever.

A small, dependency-free HTML/CSS/JavaScript prototype turned Parallel Voices into a candidate route and a recruiter route, using entirely fictional people, roles and outcomes. It let us check tasks instead of approving posters.

The candidate could inspect a grounded opportunity, save it, view an independent salary estimate only when evidence allowed one, preview a precise application packet and see a truthful synthetic receipt. The recruiter could inspect a submitted applicant separately from an opted-in prospect, ask for missing context and preview a human-written introduction without actually sending it.

The first rendered audit confirmed the problem: too much introductory explanation, both audience panels visible despite a role switch, and ornamental shapes crowding important actions. The design was expressing its own philosophy when it ought to have been helping someone do something.

What the first pass exposedTwo stages competing for attention.

The working candidate screen had around 209 displayed words at entry; the recruiter had 154. Decorative geometry intersected button rectangles in measured layouts. Those measurements did not prove blocked clicks, but they supported the usability complaint.

What the correction didOne audience. One real task.

The second pass showed one chosen workspace at a time, moved navigation before the work, reduced visible text to 76 and 59 words respectively and removed the floating decoration. Local desktop and mobile synthetic tests passed.

The word-count comparison comes from the documented October 21 local Chromium review. It is an audit observation, not a universal content-length target or proof of accessibility. The prototype stayed protected, synthetic and disconnected from real recruitment activity.

04 / Reconsideration

A cleaner dashboard still looked like a dashboard.

The simplified prototype fixed measurable clutter, but a more fundamental criticism remained: it felt flat, like another recruitment site. The next design conversation explored Relationship, Living, Editorial and a constellation-style model built around actual connections.

RelationshipPromising, but too shallow.

Better language for real people and live interactions; not yet enough useful depth when a candidate or recruiter needed to act.

LivingWarmer, but became a to-do list.

The mobile stack was readable. The experience still risked presenting recruitment as a stream of chores rather than a relationship progressing.

EditorialConsidered. Rejected.

Clear and restrained, but too distant for the active, human conversations the product was meant to support.

Constellation introduced a more interesting idea: a living network of relationships with perspective-specific categories and observable states such as Interested, Accepted and Declined. It created depth, but became less suitable as the place where an ordinary message, clarification or decision should actually happen.

Further working directions included Living Relationships and Parallel Conversations. None is claimed here as a final production design. They helped answer a more useful question: what belongs in the conversation, rather than around it?

05 / Emerging direction

Let the conversation carry the work.

The latest design direction moves away from a separate intelligence dashboard, a detached relationship graph or a stream of status cards. A person should open the relevant candidate, recruiter or company conversation and find the context, next action, evidence and history together.

That also makes future integrations with familiar messaging environments such as Slack, Teams, Discord or Google Chat easier to reason about: the product should contribute meaningful updates and inline work inside an existing conversation, not insist that everyone learns a new forum.

This is a static, fictional illustration of the proposed conversation-first model. These are not real recruiter messages, live recommendations, a working bot, or an integration with any messaging platform. “Round Up Both.” is a proposed daily conversational summary, not a deployed service.

06 / The principles that stayed

Not every iteration survived. The important rules did.

Across the visual reversals, the product's underlying commitments remained more stable than its interface. The goal is not to automate judgement of people. It is to help two sides reach an informed decision and know what they have, and have not, agreed.

Evidence over scoring

Show why a role surfaced and where experience supports a requirement. Do not manufacture a “fit” percentage or use absent CV evidence as a rejection.

Candidate-controlled visibility

Private by default. A limited professional preview requires an informed choice, and an application freezes the exact role-specific packet sent.

Human questions and decisions

“This candidate could be interesting” is an invitation to find out, not a verdict. The next action is “Introduce yourself” or “Ask for missing information”.

Honest status and salary

Received is not reviewed. An independent market estimate is not an employer offer. When reliable information isn't available, show the gap.

Intelligence that explains itself

Useful insight should tell the person what the evidence suggests changing, where, and why, within the conversation rather than on a remote analytics page.

Different roles, reciprocal respect

Candidate and recruiter don't need identical controls or mirrored slogans. They need permissions appropriate to their responsibilities and clear treatment on both sides.

Good design here should be felt in the quality of the exchange, not measured by the number of screens we've built.

07 / What remains

A direction with plenty left to prove.

There is a private, protected, client-side synthetic prototype, a detailed behavioural contract and a body of visual/voice research. The more recent constellation and conversation-led directions are design explorations, not implemented or user-tested production systems.

No real candidate service is operating. There is no real CV ingestion, recruiter database, employer matching, lawful discoverability system, messaging service, salary benchmark feed, ATS integration or hiring/billing workflow behind this paper. Future work would require accessibility and bias testing, privacy/legal review, secure identity, licensed data, verified authorisation and evidence of actual candidate/recruiter benefit.

That is precisely why I wanted to keep the earlier experiments. The wrong turns are often where the useful decisions are hiding. This page captures the route taken so far, without pretending we have already arrived.

Evidence note. The chronology reflects owner-reviewed October 20–24, 2026 discovery, an editable 48-slide behavioural/visual exploration containing 16 named visual studies, a tone-of-voice comparison, local synthetic prototype test notes and subsequent review conversations. Concept panels and message examples above are interpretive reconstructions, not original screenshots. Detailed private documents, underlying prototypes and personal/collaborator records are intentionally not published.

A practical direction, not a product claim

Better for both.
That's the whole test.

Good questions, honest evidence, no artificial certainty. The interface can keep changing. That principle shouldn't.