Platform infra

A centered goal box becomes a docked conversation, with a distinct action to build your path.

The roadmap now opens on a header card with your progress, lists each layer's chapters as rows you can open, shows where Projects and Interview prep unlock, and adds an Along the way card beside a reworked view of how your path has adapted.

The box where you describe your goal, and reply once the conversation starts, now grows with what you write, so you can see your whole message.

The learning-path composer now recovers quietly from dropped connections, waits for you when Upsoma is busy, lets you build your path with one button, and says clearly when a step fails.

When you have used your path limit, the path conversation says so plainly and offers Top up instead of a Try again button that could not work.

The locked Recommended projects and Interview prep pages now say in one line how many chapters open them, with what you will get listed beside them, and the chapter page header is simpler.

How your path has adapted now shows a small difficulty line and a short icon timeline instead of paragraphs, and every stage of a chapter uses the same buttons, panels and progress steps.

Certificates read as one sentence with a short verification link, finished paths point you to what is next, onboarding tells you how long your CV will take, and a welcome shows your plan and the launch offer.

Create a learning path through one guided conversation, with chapters chosen and ordered for your goal, background, and current readiness.

Accepted learning-path requests can now recover safely and report their saved progress while generation continues.

The footer groups every public page under seven headings, organisations can send a partnership enquiry, and CareerPalz CTL members unlock their program page with an access code that admins manage.

One side drawer now holds the Skill browser, billing with checkout, and notifications, so learners act without leaving the page they are on.

Account exports and deletion now cover saved learning paths, their conversations and generation history.

Learning-path generation can now build personalised skill and chapter outlines and save them safely as one durable result.

Learning-path discovery can now preserve context across turns and recover safely from retries and worker interruptions.

Learning-path discovery now has a native, tool-using agent definition ready for backend conversation orchestration.

Personalised path conversation transport can now move safely between legacy questions and natural agent replies.

Personalised path conversations now stay protected while native agent support rolls across backend versions.

Native agents now report reliable progress, timing and token usage without exposing learner prompts or tool data.

Agent conversations can now retain ordered, private history across restarts with safe caching and account deletion.

Personalised path generation now reserves, renews, consumes and releases path allowances through the existing billing service.

Job-posting evidence collection now stays bounded, retries fairly, and continues through every configured source without delaying the rest of the catalogue.

Personalised path execution now shares bounded background capacity with onboarding and keeps durable work ready for later agent execution.

Personalised learning paths can now preserve private conversations and complete learning outlines safely across retries.

Upsoma now collects due job postings directly into the catalogue evidence pipeline without changing the published catalogue.

Upsoma can now turn retained job-posting text into grounded role, seniority, and skill evidence for the catalogue.

Upsoma's existing model-powered decisions now use clearly named native agents and prompt modules, giving onboarding and skill evidence one consistent execution structure.

While your paths load, the Paths page shows its tabs and placeholder cards instead of a loading sentence, so nothing jumps when they arrive.

Earned certificates get their seal, code and share link; the roadmap nests chapters inside layer cards behind a segmented List / Map toggle; practice runs on one flat question card; chapter diagrams use the Evergreen palette.

Chapters name their place in the path, keep walkthroughs and diagrams a click away under the lesson, and open Ask Upsoma with prompts drawn from the chapter itself.

Review chapters before starting, learn with a focused pre-check, and continue through practice, projects and interview preparation.

Explore distinct active learners, assessment and path completions, second-path returns, acquisition, activation and retention from the admin overview.

Learners can now combine a curriculum vitae, professional links, and written background into an evidence-grounded profile, review every suggestion, and safely resume or skip the flow.

Upsoma tightened its public, protected, admin, changelog, and generation wait states so the MVP is easier to evaluate end to end and learners see useful recovery or progress instead of blank or clipped layouts.

Upsoma now presents one clearer product journey across the public site and learner app: name a goal, map the catalogue, overlay existing evidence, follow an adaptive path, and leave with readiness proof.

Upsoma now computes a live market-demand signal from real job-posting observations and puts it to work: paths teach in-demand skills sooner, your remaining gaps are ranked by what the market wants, and a new Rising card shows trending skills in your domains you don't have yet.

Admin analytics controls now sit with the analytics page identity while account actions stay isolated on the right, matching the catalogue header pattern.

The admin analytics dashboard is precomputed on the backend every five minutes and delivered with the first page render, so it opens fast even on slow connections.

The admin analytics page now uses the same connected header stack as the catalogue console, with Analytics as the single page label.

The admin analytics area shows platform usage, composer queries, user activity, sources, and learning demand signals.