Building Interviewly: How We Engineered a Multi-Provider AI Mock Interview Engine

The Challenge of AI Mock Interviews

Technical interviews are unpredictable. When candidates answer coding, architectural, or behavioral questions, an automated interviewer cannot rely on rigid scripts. It requires real-time conversational flexibility, deep syntax comprehension, and nuanced critique.

Multi-Provider Fallback Chaining

Relying on a single third-party API in production exposes your platform to rate limits, unexpected downtime, and latency spikes. In Interviewly, we engineered an automated model fallback chain:

1. Primary Route: Google Gemini Pro for ultra-fast latency and high-throughput question generation.
2. Local Failover: On-premise quantized Ollama LLM nodes when network connectivity is disrupted.
3. Advanced Evaluation: Anthropic Claude & GPT-4o for complex system design critique and code smell analysis.

Instant Competency Scoring & Feedback

Each answer is measured against standard engineering rubrics:

• Technical Accuracy (0-10)
• Communication Clarity
• Edge Case Consideration
• Alternative Approaches

The result is a transformative interview preparation tool that has already assisted thousands of developers worldwide.