Skip to content

Task.md

Latest commit

 

History

History
110 lines (109 loc) · 3.66 KB

Task.md

File metadata and controls

110 lines (109 loc) · 3.66 KB

Interview Practice Platform A web application that anyone can open and use to practice a coding interview. The user opens the web app They start a live interview session A voice-based AI interviewer: Presents a coding question (verbally and in writing) Asks follow-up questions Reacts to the candidate’s explanations The candidate: Writes code in a code editor Explains their approach verbally Iterates on the solution This is a collaborative interview The interview continues until, the candidate: Clicks End Call After ending: The code and the discussions is evaluated by an LLM The user is taken to a Results Screen What Is Explicitly Out of Scope To avoid over-engineering, do not focus on: Authentication / login User accounts Dashboards History of past interviews This is a single-session, stateless-user experience. Core Features & Flows

  1. Interview Session Flow Step 1: Start Interview Landing page has a Start Interview button Clicking it: Creates a new interview session Initializes LiveKit or Vapi Interview starts

  2. Interview Screen (Main Screen) This is the most important part of the assignment. Mandatory Sections on Screen Code Editor Panel Monaco Editor Language can be fixed (eg: JavaScript) Code state must be accessible to backend when submitting Transcription Panel Displays the agent’s transcripts Should update as and when the agent speaks. New transcripts come below and scrolls the older one. Voice Agent Panel Controls: Mute / Unmute End Call Shows speaking / listening state

  3. Voice Agent Behavior The AI interviewer should: Read out the question Ask clarifying questions Ask the candidate to explain their logic Challenge edge cases Ask complexity-related follow-ups The agent must continue the interview loop (keep asking new problems / follow up questions) until: User clicks End Call, or User explicitly says something like: “End the interview” “I’m done”

  4. Code Submission & Evaluation At the call end, you have access to the final code and the interview transcripts of both agent and the candidate, you have to process these using LLM to generate a report of that interview which is shown in the reports page.

  5. Reports Screen This screen should show: Questions Candidate’s Final Code (read-only Monaco or formatted view) LLM Feedback, clearly structured: What was done well What could be improved Missing edge cases Next steps for preparation No navigation to previous interviews is required. Tech Stack Frontend TypeScript React / Next.js Tailwind CSS ShadCN UI Monaco Editor (@monaco-editor/react or equivalent) Backend Node.js TypeScript Express MongoDB (via Mongoose) Redis AI / Media You may choose either: LiveKit (Python-based service), OR Vapi (TypeScript-based) For AI services: LLM: Groq (free tier) Speech-to-Text: Deepgram Text-to-Speech: ElevenLabs (You are free to use other providers. We recommend them because they provide free credits) Note: Deepgram and Elevenlabs provide some free credits. If any help is needed regarding credits, kindly reach out to Aditya (founder) - +91-7814520243 Submission Guidelines This assignment does NOT require live deployment. A working local setup is sufficient. Mandatory GitHub repository link (monorepo or separate repos both acceptable) README (Required) Your repository must contain a clear README explaining: How to run the frontend locally How to run the backend locally Environment variables required (with sample .env.example) Timeline You have a week to complete this assignment. You can use AI tools of your choice (ChatGPT, Claude, Cursor, Claude Code, v0.dev, etc.). Extra time can be provided if good progress is being made. Contact For any queries, you can reach out to Aditya (founder) at +91-7814520243. You can WhatsApp/Call.