Algo Hunter

Inspiration

Traditional algorithmic learning platforms spoon-feed static code or verbose text editorials. When faced with high-stakes whiteboard interviews, engineers stumble because syntax does not build intuition—spatial mental modeling does. We built Algo Hunter to bridge the intuition gap: an autonomous Socratic reasoning engine that visually simulates runtime memory frames and guides engineers with calibrated nudges rather than spoiling the answer.

What it does

Algo Hunter transforms static algorithm practice into an interactive spatial dry-run:

• Interactive Execution Tracer: Visually renders multi-pointer mechanics, indices, and 2D matrix contractions step-by-step with zero index hallucinations.
• 3-Part Socratic Reasoning: Every step breaks down the cognitive mechanics into [Observation], [The Conflict / Bottleneck], and [The Action].
• Autonomous Input Synthesis: If custom input is left blank, the engine automatically synthesizes canonical test inputs to simulate edge cases.
• Cognitive Knowledge Graph: Dynamically branches a persistent Mermaid.js DAG mapping mastery across patterns (Two Pointers, Binary Search, Matrix Simulation).
• Tavily Live Interview Radar: Real-time web search surfacing recent challenges from top tech firms (Google, Meta, Amazon) for the active pattern.
• Multi-User State Isolation: Full workspace isolation backed by FastAPI and SQLite, preserving problem histories and cognitive graphs per user.

How we built it

• Inference Engine: NVIDIA Nemotron 120B (nvidia/nemotron-3-super-120b-a12b) hosted on Nebius Token Factory, delivering low-latency structured JSON state machines.
• Backend Orchestrator: FastAPI managing algorithmic execution pipelines, workspace sessions, and database persistence.
• Database: SQLite for multi-tenant state and solved history preservation.
• Frontend Architecture: Monaco Editor for syntax and language support, custom CSS grid memory visualizer, and Mermaid.js for cognitive graphs.
• External Intelligence: Tavily Search API for real-time interview pattern radar.

Challenges we overcame

Handling complex multi-pointer and 2D matrix tracking required zero index drift. By designing a rigid AST state schema and pairing it with NVIDIA Nemotron 120B on Nebius Token Factory, we achieved zero-hallucination pointer tracking, complete non-truncated code generation, and low response latency.

What we learned

We learned that combining specialized agentic reasoning prompts with high-throughput inference creates learning interfaces far more effective than static code execution sandboxes.

What's next for Algo Hunter

• Dynamic Tree and Graph recursion stack visualizers.
• Voice-enabled real-time mock whiteboard interview mode.
• Interactive multi-user collaborative debugging rooms.

Built With

• c++
• css3
• fastapi
• html5
• mermaid.js
• monaco-editor
• nebius-token-factory
• nvidia-nemotron-120b
• python
• sqlite
• tavily-search-api