About Me

Your Name

Pranav Khetarpal

Indian Institute of Technology Delhi

Executive / Researcher — UniverseTBD

Previously: Quant Research Intern — WorldQuant, Researcher - M3RG Lab, IIT Delhi

Hey — I’m Pranav Khetarpal (IIT Delhi). I work at the intersection of LLMs / multimodal models and reasoning, with a bias toward building things that are reproducible, measurable, and useful in the wild.

Over the last couple of years I’ve been doing hands-on AI research across LLMs and multimodal models, spanning model analysis, domain adaptation, and building end-to-end systems that hold up under real constraints.

Industry angle (and a big part of my “reasoning” obsession): I was a Quantitative Research Intern at WorldQuant (Mumbai), where I built point-in-time (PIT) alternative-data pipelines powered by LLM/NLP extraction. A lot of the work was essentially reverse-engineering model behavior: turning messy text into stable, auditable features by iterating on prompt schemas, rubrics, and structured outputs until the system’s “reasoning” became reliable enough to survive distribution shift—while staying strict about provenance, leakage control, and monitoring.

  • Reasoning: grounding, tool-augmented reasoning, faithfulness/reliability, and reducing brittle shortcut behavior in LLMs/VLMs.
  • Multimodality: domain VLMs (astronomy/materials), representation learning, and robust cross-modal alignment.
  • World models: learning predictive representations and “world-model-like” structure from large-scale data; probing what emerges under different objectives (e.g., language grounding vs predictive/self-supervised training).
  • Systems for research: data pipelines, retrieval/agents, and measure-first tooling for reproducible experiments.

    • What I’m looking for right now: I’m actively exploring AI research / applied research roles (reasoning, multimodality, world-model style learning) and quantitative research roles where I can bring the same strengths—building robust pipelines, translating noisy data into signal, and stress-testing assumptions— plus a strong research discipline: careful problem framing, ablations, leakage-aware validation, and reproducible experimentation. If either aligns with your team, I’d love to talk.

    If you’re coming from an AI research lens, the best entry points are: Projects → AI/ML Research and the publication links on the projects page.
    If you’re coming from a quant lens, start here: Projects → WorldQuant and the broader Quant & Markets section.

    PS: Always happy to chat—especially around reasoning, multimodality, world models, and quant research.