An Interview Loop That Remembers Your Weak Spots

What it does

TechSpar is a technical interview coaching system that links specialized training, resume-based mock interviews, job-description prep, real-time Copilot assistance, and post-session recording reviews into one continuous loop. All modules share a single long-term learner profile, knowledge base, and weakness map; after each session, results write back to the system and shift the focus of the next round. It is built as a TypeScript monolith with a React web client and an Electron desktop app backed by a local Bun/Hono sidecar.

The interesting bit

Instead of treating AI interview prep as a prompt-engineering exercise, TechSpar structures the domain as hexagonal architecture with state machines, ports, and adapters: core business rules live in packages/core while Hono handles HTTP/SSE/WebSocket plumbing. It also runs local embeddings via Transformers.js and ONNX (Xenova/bge-m3) without dragging in Python or PyTorch, and uses the SM-2 algorithm to schedule your reviews like a spaced-repetition deck.

Key highlights

• Shared long-term memory: a unified profile, knowledge base, and weak-point tracker feeds every module from specialized drills to JD-specific prep.
• Real-time Copilot: ASR transcription, follow-up prediction, answer suggestions, risk warnings, and optional voiceprint recognition via Tencent Cloud VPR.
• Offline-friendly local embeddings: generates vectors using Transformers.js + ONNX, caching models locally with no Python dependencies.
• Desktop and web: Electron 43 client with a compiled Bun sidecar, plus a Docker-ready web stack.
• Data portability: exports include tar checksum validation, atomic restore with timestamped backups, and sensitive credentials are excluded by default.

Caveats

• Licensed under CC BY-NC 4.0, so commercial use is off the table.
• Desktop releases as of v0.3.1 lack commercial code signing; macOS and Windows may flag the installer.
• The legacy Python/FastAPI branch is frozen and the README warns against mixing its data directory with the current TypeScript backend.

Verdict

Worth a look if you want a structured, self-hosted interview coach that adapts to your actual gaps over time. Skip it if you need a managed SaaS, a mobile app, or a commercially licensable tool.