PromptRank

PromptRank is an AI-collaboration assessment platform that simulates realistic work tasks with an AI assistant and evaluates how candidates use it. Instead of scoring memorized coding puzzles, PromptRank looks at whether candidates verify AI output, catch confident mistakes, calibrate trust, and work effectively with AI under realistic conditions.

The platform has two views.

• The Candidate View walks a job-seeker through a live, realistic work simulation with an AI assistant, then delivers a personalized report with scores, feedback, and upskilling recommendations.
• The Employer/Institution View gives training providers and hiring teams a dashboard of aggregate cohort results - pass rates, average scores, and drill-down into any individual candidate's full assessment report.

Getting Started

1. Clone the repo.
2. Run npm install.
3. Copy .env.example to .env and add a real OpenAI API key.
4. Run npm run dev.

Tech Stack

• React
• TypeScript
• Vite
• Tailwind
• OpenAI API (gpt-4o-mini)
• React Router
• Lucide React

Live Demo

Live demo: https://prompt-rank-rho.vercel.app/

Slide Deck

Website views: https://canva.link/fz7pr2n0x4g2lbw

Agent Architecture

PromptRank runs five core agents per assessment session. The first five run in series, since each depends on the previous agent's output; the final three run in parallel once grading completes, since they all consume the same finished score independently.

Serial pipeline:

1. Intake & Calibration Agent - reads the candidate's resume and job description, matches them to an assessment track, and generates a customized brief.
2. Scenario Architect - supplies the scripted task template and turn-by-turn correct/flawed schedule for that track.
3. Dynamic Simulator Agent - runs the live chat with the candidate, following the scenario's turn schedule (reliable by default, deliberately flawed at a specific pre-planned turn).
4. Behavioral Observer Agent - analyzes the completed transcript against the turn schedule to flag exactly when flaws occurred and how the candidate responded.
5. Executive Grader Agent - consumes the Observer's findings to compute structured rubric scores and written feedback.

Parallel finish (triggered once the Grader completes):

• Peer Calibration - computes the candidate's percentile against benchmark data.
• Ecosystem Connector - maps weak rubric categories to MySkillsFuture-aligned course recommendations.
• Dual Report Synthesis - renders the same score data into both the Candidate Report and Institution Report views.

Project Structure

• src/pages/: route-level pages for Login, Profile, Start Session, Session, Processing, Report, Skills Upgrade, Session History, My Profile, Pricing, and the Employer Dashboard.
• src/components/: shared UI components such as Button, Card, ProgressStrip, ChatBubble, Navbar, and Badge.
• src/agents/: OpenAI-backed agent wrappers for intake summaries, simulation responses, behavioral observation, and grading.
• src/data/: hardcoded assessment scenarios, peer score benchmarks, course recommendations, and past session records.
• src/context/: app-wide session state managed with React context and a reducer.
• src/types/: shared TypeScript interfaces for scenarios, transcripts, observer output, grader output, and rubric scores.

Environment

• .env.example contains VITE_OPENAI_API_KEY=your_openai_api_key_here.
• .env is intentionally gitignored and should contain the real key for local demos.