Roadmap

Context-aware follow-up prompts grounded in the conversation so far, so the assistant guides recruiters toward the evidence that matters to them.

Smoother barge-in handling and explicit clarification when audio is unclear, instead of guessing — voice should feel like a conversation, not a kiosk.

A "request your own agent access" flow that issues a per-account token tied to a verified email, so any MCP-capable agent can query this CV directly — moving beyond today's single shared, rate-limited token.

Promote the dual ONNX prompt-injection classifiers (already CI-tested and benchmarked, fail-open) from available-behind-a-flag to live in production after false-positive and latency validation.

Richer gap analysis that explains how specific experience maps to a pasted job description, with grounded citations rather than a single score.

Unify the rendered case-study copy and the RAG grounding so a single source drives both — eliminating drift between what the page shows and what the assistant says.

An embedded, lazy-loaded mini-demo that lets a visitor watch a retrieval query resolve on the page itself, without launching the full chat.

Ongoing polish of mobile interactions, focus management, and motion so the experience stays fast and accessible on every device.