Research Fellowship: Agent Intelligence & Evaluation New

NetVerks Inc.; Delhi; On-site; INR 508,000 - 608,000; Full time

Job summary

ixigo seeks a research-focused fellow to advance self-healing voice agents for enterprise support. You will design evaluation frameworks, build end-to-end observability, and drive self-improvement loops to trace failures across ASR, LLM, and TTS components, aligning metrics with real-world outcomes.

During a four-month fellowship, you’ll access real enterprise conversation data under governance, receive mentorship, and potentially co-author on papers.

Qualifications

• PhD students in ML/NLP/speech are preferred; exceptional MS students or researchers with publications are welcome.
• Proficient in Python and familiar with at least one: Whisper, Conformer, or LLM tooling.
• Experience with evaluation methodology, dataset synthesis, or interpretability is a plus.

Responsibilities

• Design audio-native metrics and datasets for voice-agent evaluation.
• Build end-to-end observability linking audio, STT, LLM, tools, and TTS across conversations.
• Develop self-improvement loops by mining production traces and creating targeted fine-tuning data.

Skills

Python

Speech models (Whisper, Conformer)

LLM tool-use & agent frameworks

Observability stacks (OpenTelemetry, L

Education

PhD students in ML/NLP/speech

Tools

OpenTelemetry

Langfuse

Arize

Hamming

Job description

We’re building self-healing voice agents for enterprise customer support within ixigo. The system has to know when it’s failing, why it’s failing, and how to fix itself before a human notices. This fellowship sits at the intelligence layer behind that work.

Voice agents fail in ways traditional software doesn't. An ASR confidence drop on a regional accent misfires a tool call, an LLM hallucinates a policy because upstream latency broke turn-taking, and support teams roll these agents back within a week without anyone able to explain what went wrong.

What you'll work on

• Evaluation frameworks. Text-only evals miss most of what matters in voice: barge-in, prosody, latency-induced errors, cross-turn context loss. You'll design audio-native metrics, generate adversarial conversational datasets across accents and edge cases, and build LLM-as-judge rubrics for task completion, empathy, and recovery from tool failures.
• End-to-end observability. Tracing a failed interaction means correlating audio packets, STT hypotheses, LLM reasoning traces, tool calls, and TTS output back to a single conversation ID. You'll help shape the schema and analysis layer that makes cascade failures visible across the stack.
• Self-improvement systems. Once you can measure and trace, the interesting work is closing the loop: mining production traces for failure patterns, generating targeted fine-tuning data or prompt updates, and validating that fixes hold under adversarial replay.

What you'll get

₹50,000/month for the 4-month term, access to real enterprise conversation data under proper governance, mentorship on the research and shipping sides, co-authorship on papers that come out of the work, and a system in production running on top of what you build.

Candidates are responsible for safeguarding sensitive company data against unauthorized access, use, or disclosure, and for reporting any suspected security incidents in line with the organization's ISMS (Information Security Management System) policies and procedures.

Requirements

Who we're looking for

Someone who cares about the research questions for their own sake, and equally cares whether the work ships. Papers at Interspeech, ACL, NeurIPS, or EMNLP on speech, dialogue systems, agent evaluation, or human-AI interaction are directly relevant.

Comfortable in Python, and familiar with at least one of: speech models (Whisper, Conformer variants), LLM tool-use and agent frameworks, or observability stacks (OpenTelemetry, Langfuse, Arize, Hamming). Current PhD students in ML, NLP, or speech are the strong default; exceptional MS students or research engineers with a publication track record are welcome to apply.

Nice to have

Prior work on evaluation methodology, dataset synthesis, or interpretability. Experience with real-time systems, telephony, or streaming pipelines. A blog, repo, or workshop paper that shows how you think in public.