From Dusk Till Dawn: Waymark
Career direction for people who do not yet know what they want, built from the real job market rather than a personality quiz. Built overnight by a team of three at the From Dusk Till Dawn hackathon in Prague, then hardened for a public demo.
// evidence
- build
- 21:00 – 7:14
- team
- three people
- my part
- intake + roadmap
- track
- Deep Research
// The event
From Dusk Till Dawn #01, run by the Agents 0.0.7 community at Etnetera in Prague, for teams of two or three. Building started at nine in the evening and the code froze at sunrise, 7:14. The submission was the codebase and a two-minute demo video, with presentations at half past nine.
We took the Social Media Deep Research track: an agent that researches a person or an organisation for a given goal. Waymark points that research at the job market around one person’s goal. No placing.
// The problem it picks
Career advice for someone who is not sure yet is mostly vibes: personality quizzes, top-ten job lists, and match percentages nobody can explain. Job ads, meanwhile, already say exactly what employers pay for, in their own words.
// How it works
- Intake from job ads
- Instead of a quiz, the seeker reacts to everyday tasks taken from real job ads. Each next card is the one that best separates the paths the engine is least sure about, and it stops once the top three are clear. It can also be talked through by voice.
- Live market research
- An agent reads job boards, company pages, registries and salary sources for the seeker’s locations. Every requirement is extracted with a verbatim quote, and if the quote is not in the fetched page, the claim is rejected.
- Up to three paths
- Open roles, salary ladders where a source gives one, companies hiring now — each with linked evidence. No match percentage and no hiring probability.
- A sourced roadmap
- One chosen path becomes learning chapters, and each skill shows how many current ads actually ask for it.
// My part
The seeker side: the guided intake and adaptive task cards, interview and CV reading, validation, the roadmap planner, privacy controls, the ElevenLabs voice integration, wiring the live UI, and the deployment.
// What it is not yet
A hackathon project, and the repository says so. The task deck still runs on stub occupations, the API sits behind one shared key instead of real accounts, and recommendation quality has not had a human evaluation.
So the public demo runs on sample data only. Anyone can click through the whole product, nothing they type leaves the browser, and no paid model is ever called.
// the product
