Research Fellowship: Agent Intelligence & Evaluation

Design and develop audio-native evaluation frameworks and adversarial conversational datasets to assess voice agent performance.

Contribute to end-to-end observability by shaping analysis tools correlating multi-stage voice agent failures across system components.

Build self-improvement systems by mining production traces for failures, generating fine-tuning data, and validating fixes under adversarial conditions.

Proven research contributions with papers at conferences such as Interspeech, ACL, NeurIPS, or EMNLP in relevant domains (speech, dialogue systems, agent evaluation, or human-AI interaction).

Proficiency in Python and familiarity with at least one of: speech models (e.g., Whisper, Conformer variants), LLM tool-use and agent frameworks, or observability stacks (e.g., OpenTelemetry, Langfuse).

Currently enrolled PhD student in ML, NLP, or speech or exceptional MS students/research engineers with publication records.

Work Experience Required: Not explicitly mentioned in the JD.

Strong research background with a proven publication record in speech, dialogue systems, or human-AI interaction targeting top ML or NLP conferences.

Technical expertise bridging speech models, language model agents, and observability pipelines to handle complex end-to-end voice system failures.

Ability to work on both theoretical research questions and practical system shipping within a short-term (4-month) fellowship timeline.