ip Indraneil Paul

Applied Scientist PhD Intern at Amazon (AWS), Berlin

Indraneil Paul

PhD Researcher, Language Models & Code

I'm a PhD researcher at the UKP Lab, TU Darmstadt, advised by Iryna Gurevych and Goran GlavaΕ‘. I work on the mid- and post-training of language models, with an emphasis on agentic coding and tool use.

My longer-term aim is to extend LMs to long-horizon operation β€” computer use, recursive workflows β€” by improving how they reason, offload computation, and learn from environment feedback. Alongside this, I study preference learning and verifiers that push code models along non-functional axes like security and efficiency, and I care about the whole pre-training stack: data curation, context-length extension, modularity, and reinforcement learning.

Previously I was an Applied Scientist at Amazon, and before that a dual-degree student at IIIT Hyderabad. I've contributed to several open code-LM releases, including StarCoder2 and BigCodeBench.

Portrait of Indraneil Paul

πŸ”¬Research

Three threads I keep pulling on.

  1. Agentic coding & tool use

    Teaching code models to operate over long horizons β€” calling tools, offloading computation, and learning from execution and environment feedback rather than static text alone.

    OctoLongBigCodeBenchStarCoder 2 and The Stack v2

  2. Verifiers & preference learning

    What actually makes RLVR and reward models work for code, and how to score generations along non-functional axes β€” correctness, security, efficiency β€” across languages and criteria.

    AletheiaThemis

  3. Pre-training efficiency & grounding

    Getting more out of code-LM pre-training through obfuscation and intermediate-representation grounding, multilingual transfer, and modular / parameter-efficient methods.

    ObscuraCoderIRCoderAdapters

πŸ“šPublications

Selected publications.

  1. 2027

    OctoLong: Mid-Training on Cross-Repository Code Contexts Enhances Long-Context Modeling β˜…

    Indraneil Paul et al.

    NAACL Under Review
    Abstract PDF
  2. 2026

    Themis: Training Robust Multilingual Code Reward Models for Flexible Multi-Criteria Scoring

    Indraneil Paul et al.

    TMLR Under Review
    Abstract PDF
  3. 2026

    Aletheia: What Makes RLVR for Code Verifiers Tick? β˜…

    Vatsal Venkatkrishna et al. (incl. Indraneil Paul)

    TMLR
    Abstract PDF
  4. 2026

    AICD Bench: A Challenging Benchmark for AI-Generated Code Detection

    Daniil Orel et al. (incl. Indraneil Paul)

    EACL, Rabat
    Slides Abstract PDF
  5. 2025

    Droid: A Resource Suite for AI-Generated Code Detection

    Daniil Orel et al. (incl. Indraneil Paul)

    EMNLP, Suzhou
    Slides Abstract PDF
  6. 2025

    ObscuraCoder: Powering Efficient Code LM Pre-Training via Obfuscation Grounding β˜…

    Indraneil Paul et al.

    ICLR, Singapore Poster
    Slides Abstract PDF
  7. 2025

    BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions β˜…

    Terry Yue Zhuo et al. (incl. Indraneil Paul)

    ICLR, Singapore Oral
    Abstract PDF Project
  8. 2024

    IRCoder: Intermediate Representations Make Language Models Robust Multilingual Code Generators β˜…

    Indraneil Paul et al.

    ACL, Bangkok Oral Outstanding Paper
    Slides Abstract PDF
  9. 2024

    StarCoder 2 and The Stack v2: The Next Generation β˜…

    Anton Lozhkov et al. (incl. Indraneil Paul)

    TMLR
    Slides Abstract PDF
  10. 2023

    Adapters: A Unified Library for Parameter-Efficient and Modular Transfer Learning

    Clifton Poth et al. (incl. Indraneil Paul)

    EMNLP System Demonstrations, Singapore
    Demo Abstract PDF
  11. 2022

    Sub-Task Imputation via Self-Labelling to Train Image Moderation Models on Sparse Noisy Data

    Indraneil Paul et al.

    CIKM, Atlanta Oral
    Slides Abstract PDF

πŸ“°News

Recent updates.

  1. Co-organizing the SemEval 2026 Task on GenAI Code Detection & Attribution.

  2. AICD Bench accepted to EACL 2026 (Rabat).

  3. Aletheia, on what makes RLVR for code verifiers tick, accepted at TMLR.

  4. Droid, a resource suite for AI-generated code detection, accepted to EMNLP 2025 (Suzhou).

  5. Started an Applied Scientist PhD internship at Amazon (AWS) in Berlin, working on RL for cloud tool-calling in Amazon Q Developer.

  6. BigCodeBench (Oral) and ObscuraCoder (Poster) presented at ICLR 2025 (Singapore).

  7. IRCoder received an Outstanding Paper Award at ACL 2024 (Bangkok).

  8. StarCoder 2 and The Stack v2 released.

βœ‰οΈContact

Get in touch.

I'm always glad to talk about code models, verifiers, and long-horizon agents β€” or to hear about roles and collaborations. Reach me by email, or find me on the profiles below.

indraneil.paul@gmail.com View CV Download CV (PDF)