Skip to content

TECHNICAL_FLOW.md

Latest commit

 

History

History
61 lines (46 loc) · 3.58 KB

TECHNICAL_FLOW.md

File metadata and controls

61 lines (46 loc) · 3.58 KB

🎮 Technical Simulation & Session Flow: The Master Blueprint

This document outlines the architecture for the Technical Coding Rounds, AI Judging System, and the Progression Game Loop for InterviewAI.

🏗️ 1. The "LeetCode" Intelligence (Problem Sourcing)

Instead of generic questions, the system generates company-specific technical challenges.

  • Source: Integrated with the Agentic Discovery System.
  • Contextual Generation:
    • If the company is Groq, the problem focuses on Hardware-aware software optimization or Parallel Array processing.
    • If the company is OpenAI, the focus shifts to Transformer attention mechanisms or Vector search optimizations.
  • Format: The AI provides a detailed Problem Statement, Sample Inputs/Outputs, and a "Bar" (Expert-level criteria).

🖥️ 2. The Coding Sandbox & AI Judge (Execution Layer)

We utilize a high-performance simulation model for code evaluation without the overhead of heavy containerization.

  • The Sandbox (Frontend): Implemented using Monaco Editor (VS Code Engine) for a premium, developer-first experience.
  • Option B: The AI Judge (Selected):
    • Mechanism: The user's code is sent to the Llama-3/Gemini backend.
    • Verification: The AI "Dry Runs" the code against the hidden test cases.
    • Evaluation Metrics:
      1. Correctness: Does it solve the logical core?
      2. Complexity: Is the Time/Space complexity optimal for that specific company’s standard?
      3. Code Quality: Readability, variable naming, and "Role Readiness" (e.g., SDE-1 vs Senior).

📈 3. The "Level-Up" Progression (The Game Loop)

The interview follows a strict Gated Entry flow to simulate a real high-stakes hiring process.

  • Round Locking: Round 1 (Technical) must be cleared with a score of >70% to unlock Round 2.
  • Session State: Managed via Neon Cloud (PostgreSQL) tracking current_round, round_status, and accumulated_feedback.
  • The Gatekeeper: A backend service that checks the AI Judge's verdict before updating the user's progress.

🧬 4. Role-Specific Logic (AI/ML vs SDE)

The "Bar" for passing varies significantly based on the user's target role:

Role Category Technical Focus AI Judge Weighting
SDE / Backend DS & Algorithms, System Latency Efficiency, Edge Cases
AI/ML Engineer Array Math (NumPy), Optimization Mathematical Logic, Data flow
GenAI Developer Prompt patterns, RAG logic, Vector DBs Pattern recognition, Semantic understanding

🛠️ 5. The "Fail" Branch (The Learning Roadmap)

Failure is transformed into an opportunity for growth.

  • Post-Mortem: If a user fails a round, the Learning Roadmap Agent is triggered.
  • Personalization: It analyzes the specific mistakes in the code and generates a 7-Day Study Plan targeted at the user's weak points.
  • Infinite Loop: Once the user completes the roadmap, they can "Re-attempt" the high-fidelity simulation.

✅ Implementation Status Note

Previously: This document contained a "Loophole Warning" blocking implementation of the Coding Round system.

Resolved (February 2026): The loophole has been addressed. The full Coding Round solution — including Whiteboard Mode, AI Dry Run, Tiered Hint System, JSON Learning Ledger, and the Persona Architecture — is now designed and documented in:

📄 CODING_ROUND_DESIGN.md

Implementation begins in the next sprint (Phase 6).