SpeakCode | nwHacks 2026
"Coding is only half the battle."
In top-tier technical interviews, a passing test suite isn't enough. Engineers are expected to articulate complex logic, justify trade-offs, and maintain composure—all while the clock is ticking. SpeakCode is a multimodal practice platform that evaluates not just the code you write, but the way you explain it.
The Challenge
Modern technical screenings have a "Communication Gap." Platforms like LeetCode focus entirely on syntax and Big O, ignoring the verbal reasoning that actually gets you the job. Companies look for:
- Logic Articulation: The ability to explain why a Hashmap was chosen over a TreeMap.
- Technical Clarity: Explaining data structures clearly without losing focus on the implementation.
- Acoustic Confidence: Managing pacing, tone, and filler words under high-stress conditions.
How it Works
SpeakCode uses a specialized AI pipeline to analyze your performance across two dimensions simultaneously:
1. The Content
- Whisper Large-v3: Transcribes your verbal reasoning with high fidelity, even with heavy technical jargon.
- GPT-4 Intelligence: Acts as a "Critical Interviewer" to grade your logic, algorithmic accuracy, and code efficiency against FAANG standards.
2. The Delivery
- Signal Processing: We use
librosaandparselmouthto extract acoustic features like pitch variance (confidence), speech rate (pacing), and filler word density (clarity).
Core Features
- Live Pressure Simulation: A minimal, high-stakes IDE designed to mimic real-world technical screenings.
- Multimodal Feedback: Get a holistic dashboard showing your Communication Score alongside your Code Correctness.
- Logic vs. Syntax Analysis: Our AI distinguishes between a "syntax error" and a "logic gap," giving you specific feedback on where your explanation failed.
- Acoustic Metrics: Visualize your speech patterns to identify where you sound hesitant or rushed.
Technical Architecture
| Layer | Technologies |
|---|---|
| Frontend | React 19, Vite, Tailwind CSS, Monaco Editor |
| Backend | Python (Flask), Node.js, WebSockets |
| AI/ML | OpenAI GPT-4, Whisper-1 (Speech-to-Text) |
| Signal Processing | Librosa, Parselmouth (Praat-based pitch analysis) |
| Infrastructure | Docker, Fly.io (Machine-based code execution) |
Built With
- node.js
- passport-js

Log in or sign up for Devpost to join the conversation.