DEV Community

Hari Haran
Hari Haran

Posted on

HumanFit: Hire for What People Can Actually Do, Not What Their CV Claims

This is a submission for the [CITY OF
Rethinking Recruitment + https://harishockings--humanfit-web.us-east.modal.run/].

What I Built

HumanFit is an AI-assisted, skills-first recruitment platform. Instead of starting with a CV, it starts with evidence.

A CV tells you what someone claims to know. In an age where AI can write the whole application, that is no longer enough. HumanFit lets people prove their skills through practical, realistic challenges, and gives employers evidence they can trust.

One platform, three perspectives:

  • Espoo (organizer): runs recruitment events, tracks the pipeline, and reviews submissions before anything is issued.
  • Recruiters: define the roles and competencies they need, and create practical challenges in the AI Digital Twin Studio.
  • Participants: receive invitations, complete project-based tasks, and earn a Skill Passport.

Demo


Live app: [LINK, or remove this line]

How It Works

  1. Brief: a recruiter defines a role and the practical skills it needs.
  2. Task: HumanFit generates a realistic challenge with a transparent rubric.
  3. Digital Twin: the participant works through a 3D simulated scenario. Example: an elderly-care unit, a norovirus outbreak, two assistants, and three hours until dinner service. They set a quarantine perimeter, prioritise disinfection zones, allocate limited resources, and record a verbal handover.
  4. AI assessment: the submission is scored against the rubric, with strengths and evidence gaps called out, plus targeted interview questions such as "Why did you prioritise this zone first?"
  5. Human review: an organizer checks the evidence, adjusts scores, adds comments, and approves.
  6. Skill Passport: the participant gets a record of what they demonstrated, and recruiters can match on it.

AI assists. Humans decide. The AI never makes the final hiring decision.

Tech Stack

  • FastAPI: API endpoints for the app, AI workflows, uploads, and interviews.
  • Tiger Data: hosted PostgreSQL-compatible storage for app records, evaluation data, and audit events.
  • Modal: deploys the frontend and API as separate serverless apps.
  • Google Gemini 3.5 Flash: generates hiring tasks and interview questions, evaluates submissions and interview answers, and transcribes handover audio (Gemini 2.5 Flash as fallback).
  • ElevenLabs: speaks interview questions aloud and transcribes candidate responses in English, Finnish, and Swedish (browser speech as fallback).
  • Cloudinary: stores profile pictures and candidate submission files (local storage as fallback).
  • Three.js: the interactive 3D digital twin.
  • React + TypeScript + Tailwind: the frontend.

Code

Challenges I Ran Into

  • [e.g. keeping AI scoring transparent and tied to a visible rubric]
  • [e.g. building a 3D scenario that runs smoothly in the browser]
  • [e.g. supporting interviews in three languages]

What I Learned

[2–3 sentences in your own words.]

What's Next

  • [e.g. more scenario types beyond care work]
  • [e.g. pilot with a real recruitment event]
  • [e.g. deeper audit trail and fairness reporting]

Honest Status

HumanFit is a hackathon prototype. The demo uses synthetic data, the Digital Twin is a simulated assessment and not a clinical certification, and nothing here guarantees a hiring outcome.


What would you want to see in a Skill Passport? Tell me in the comments.

Built by Hariharan / TEAM, with DEV Hari.

Top comments (0)