Overview & AI Agents
interviewPrepper is a sophisticated AI-powered learning platform that takes a job posting URL and synthesizes a completely personalized 30-day study roadmap. By leveraging autonomous AI agents to scrape, comprehend, and structure curriculum data, it directly attacks the overwhelming "what do I study" phase of modern interview preparation.
LangGraph Orchestration
The core intelligence isn't a single LLM prompt—it's a multi-step directed acyclic graph (DAG) executed via LangGraph. This ensures accuracy and allows for dynamic fallback mechanisms during execution.
Scrapes Job Description] --> QualityCheck{Data Quality Check} QualityCheck -->|Failed| WebSearch[Tavily Search Agent
Find Role Info manually] QualityCheck -->|Passed| Analyzer WebSearch --> Analyzer Analyzer[Analysis Node
Extract core skills] --> Planner Planner[Planner Node
Draft 30-day curriculum] --> Reviewer Reviewer[Reviewer Node
Ensure actionable tasks] --> UI UI[React App UI] --> Revisions Revisions[Chat Agent
Threaded revisions] --> UI style User fill:#ffe156,stroke:#1a1a1a,stroke-width:2px,color:#1a1a1a style QualityCheck fill:#fafafa,stroke:#1a1a1a,stroke-width:2px,color:#1a1a1a
Platform Features
To avoid vendor lock-in and optimize costs/reasoning capabilities, the backend abstracts the Language Model layer through LangChain. It actively supports routing prompts through OpenAI (GPT models), Google Gemini, or OpenRouter via strict `.env` environment isolation.
Once the initial 30-day plan is generated, users aren't stuck with a static document. The application maintains persistent conversation threads via SQLite. Users can chat directly with the AI within the context of their specific job application to expand on topics or request resource links.
Full JWT-based state management ensures that learning roadmaps and private conversation threads are strictly isolated per-user.
Core Technology
- Backend AI: FastAPI acts as the primary API host, while LangChain and LangGraph execute the multi-step agent reasoning loops.
- External Services: Firecrawl is utilized for resilient bypassing of bot-protection on popular job boards, and Tavily provides intelligent retrieval-augmented generation (RAG) when a job posting lacks sufficient technical depth.
- Frontend: React 18, utilizing Vite for rapid HMR development, and styled with custom standard CSS built on Neobrutalism. React Markdown securely renders the AI's curriculum outputs natively into the DOM.