TechSpar addresses the statelessness of typical AI interview tools by implementing a persistent, long-term memory system. It builds detailed user profiles tracking weaknesses and progress, enabling continuously evolving, personalized technical interview training. This system targets developers preparing for technical roles seeking adaptive practice beyond one-off Q&A.

How It Works

Its core innovation is a personalized question engine fusing session context (resume, JD, knowledge base), topic mastery (weaknesses, trajectory), and a global profile (strengths, thinking style) to generate relevant, targeted questions. TechSpar employs a continuous training loop: each session involves per-question evaluation, weakness extraction, mastery updates, and SM-2 spaced repetition scheduling. This cyclical process ensures subsequent training is informed by past performance. The architecture uses a React 19 frontend, FastAPI/LangGraph backend, SQLite with vector embeddings, and integrates with OpenAI-compatible LLMs and optional services (DashScope ASR, Qiniu OSS).

Quick Start & Requirements

Setup is via Docker (docker compose up --build) or manual installation. Key requirements include an OpenAI-compatible LLM and Embedding API. Optional features like audio transcription need DashScope API keys; file storage can use Qiniu OSS. Environment variables in .env manage API keys and models. A live demo is at https://aari.top/ (use admin@techspar.local / admin123). See English README for details.

Highlighted Details

• Persistent user profiling and long-term growth tracking.
• Dynamic question generation targeting specific weaknesses.
• SM-2 spaced repetition for review scheduling.
• Simulated interviews based on resumes (self-intro, technical, projects).
• Job Description (JD) targeted preparation.
• Audio/text interview recording analysis with structured feedback.

Maintenance & Community

The README provides no details on specific contributors, sponsorships, or community channels (e.g., Discord, Slack). The project appears publicly hosted on GitHub.

Licensing & Compatibility

Licensed under the MIT license, permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

The online demo warns against uploading sensitive personal information, indicating it's unsuitable for confidential data. Functionality depends on external LLM/Embedding services, potentially incurring costs. Optional features require additional setup.