Improve interviewer prompts for a more fluid, natural conversation

[LlanerasJ]

Problem

The interview conversation can feel stiff/robotic rather than a natural back-and-forth. The interviewer prompts and response behavior need tuning so the exchange flows like a real interview.

Goal

Make the AI interviewer feel more fluid, human, and responsive across both modes.

Scope

• Text interviewer prompt (lib/prompts.ts -> interviewerSystemPrompt):
  • More natural acknowledgements and transitions between questions.
  • Smarter follow-ups that react to the content of the answer (probe specifics, ask "why", chase missing detail) instead of generic next-questions.
  • Better pacing - vary question length, avoid repetitive phrasing, don't over-explain.
  • Handle short/uncertain answers gracefully (encourage, rephrase) without breaking character.
• Avatar interviewer context (lib/tavus.ts -> buildInterviewerContext):
  • Keep the brief one-sentence reactions natural and varied (avoid sounding scripted).
  • Tune for spoken delivery (shorter sentences, conversational tone).
• Keep the separation: still NO formal scoring/critique during the live interview (that stays with the evaluator).

Ideas

• Light persona variation (warm vs. tough interviewer) as an option.
• Few-shot examples of strong follow-ups in the prompt.
• Tune temperature / response length for the interviewer model.

Done when

• Interviews (text + avatar) feel noticeably more natural and responsive in manual testing, with follow-ups that clearly react to what the candidate said.