Research Scientist - AI / ML Research

Roadzen, Inc.; Delhi; On-site; INR 1,500,000 - 3,500,000; Full time

Job summary

Roadzen, Inc. seeks a Research Scientist to advance AI/ML across LLMs, CV, voice/speech, and agentic AI. You will design and run experiments on architectures, training, and evaluation, collaborating with engineers to move ideas from hypothesis to shipped capabilities.

You will explore LLM reasoning, alignment, and agentic planning, or CV/vision and speech pipelines, pursuing research that translates into real product improvements.

Qualifications

• M.Tech or equivalent in ML/AI/CS or related field.
• Proven research quality or impact in LLMs, CV, speech, or agentic AI.
• Hands-on experience training/fine-tuning/evaluating DL models.
• Strong experimental design and statistical reasoning.

Responsibilities

• Design, run, and analyse experiments across AI domains: LLMs, CV, Voice/Speech, or Agentic AI.
• For LLM/Agentic projects: research training, fine-tuning, and alignment techniques; develop agentic capabilities.
• For Computer Vision: research and develop model architectures for detection, classification, segmentation, or multimodal understanding.
• For Voice/Speech: research models for speech recognition, synthesis, or conversational pipelines.
• Build evaluation frameworks and benchmarks to measure model performance beyond leaderboards.
• Investigate failure modes and propose mitigations.
• Collaborate with engineering to translate research into production training pipelines.
• Stay current with research literature; represent the team through reports or publications where appropriate.
• Design and run ablations to isolate impact of architectural, data, and training changes.
• Partner with infra teams to scale experiments across compute.

Skills

Research experience

Python programming

Deep learning

DL frameworks

Experiment design

Education

M.Tech or equivalent

Tools

PyTorch

TensorFlow

JAX

Job description

We're looking for a Research Scientist to advance the state of the art across our AI/ML portfolio spanning Large Language Models, Computer Vision, Voice/Speech systems, and Agentic AI. You'll design and run experiments on model architecture, training, and evaluation, working closely with engineers to move ideas from hypothesis to shipped capability. Depending on team needs and your interests, you may work on LLM reasoning and agentic behaviour, Computer Vision models, conversational voice bots, or a mix of these. This role is well suited to someone who enjoys deep technical research but is equally motivated by seeing that research translate into real product improvements.

Responsibilities

• Design, run, and analyse experiments across one or more AI domains: LLMs, Computer Vision, Voice/Speech, or Agentic AI depending on project needs.
• For LLM/Agentic projects: research training, fine-tuning, and alignment techniques (e. g., pretraining, RLHF/RLAIF, instruction tuning, preference optimisation), and develop agentic capabilities such as planning, tool use, and multi-step reasoning.
• For Computer Vision projects: research and develop model architectures for tasks such as detection, classification, segmentation, or multimodal vision-language understanding.
• For Voice/Speech projects: research and develop models for speech recognition, synthesis, or conversational voice bot pipelines (ASR, TTS, dialogue management).
• Build evaluation frameworks and benchmarks to rigorously measure model performance beyond standard leaderboards, tailored to the relevant domain.
• Investigate failure modes (e. g., reasoning errors, misclassifications, transcription errors, tool-use hallucination) and propose mitigations.
• Collaborate with engineering teams to translate research findings into production training pipelines and deployed model capabilities.
• Stay current with, and contribute to, the research literature relevant to your focus area(s); represent the team's work through internal reports, papers, or external publication where appropriate.
• Design and run ablation studies to isolate the impact of architectural, data, and training changes.
• Partner with infrastructure teams to scale experiments efficiently across available compute.

Requirements

• M. Tech (or equivalent advanced degree) in Machine Learning, AI, Computer Science, Electrical Engineering, or a related field.
• Strong research background or demonstrated project impact in at least one of: LLMs/NLP, Computer Vision, Speech/Voice AI, or Agentic systems.
• Hands‑on experience training, fine‑tuning, or evaluating deep learning models relevant to your area of focus.
• Strong software engineering skills in Python and experience with deep learning frameworks (PyTorch, TensorFlow, or JAX).
• Ability to design rigorous experiments and evaluations, with a strong grasp of statistical reasoning and eval methodology.
• Comfort working in a fast‑moving research environment with evolving priorities, incomplete information, and cross‑domain collaboration.

Nice To Have

• Experience across multiple AI domains (e. g., both LLMs and Computer Vision, or Voice and Agentic systems).
• Experience with large-scale distributed training (multi-node, multi‑GPU) and performance optimisation.
• Contributions to open‑source ML/AI frameworks or published research.
• Experience building evaluation suites for reasoning, coding, vision, or speech tasks.
• Prior work on model safety, interpretability, or robustness.
• Track record of translating research prototypes into production-quality systems.