( Internships )

Experience.

Research and engineering roles spanning long-context LLMs, distributed inference, and computer vision.

Internships
4
Resume
PDF ↗

( 01 — Timeline )

Where I’ve worked

  1. September 2026 – Present
    Inference Optimisation

    ML Engineer Intern

    Simplismart

    • Working on LLM inference optimisation for production serving, with research spanning both the pipeline level and the kernel level.
    • Pipeline level: profiling the end-to-end serving path to locate throughput and latency bottlenecks, and researching optimisations across request scheduling, batching, and memory handling.
    • Kernel level: researching and benchmarking GPU kernels for core inference operators such as attention, aiming to cut per-token latency on deployed models.

    LLM inference / GPU kernels / Profiling / Model serving

  2. August 2026 – September 2026
    Long-Document Generation

    AI Research Intern

    Lexsi Labs

    • Worked on long-document generation and led the research project BRIEF (Bounded Regeneration In Exact Form), which regenerates long legal and financial documents clause by clause from short structural briefs derived from a reference graph.
    • Kept cross-reference fidelity near 0.98 across a controlled length sweep with a near-constant ~86-token maximum prompt, where one-pass generation exceeds 20K tokens per call at 1,200 clauses.
    • Improved extractor-verifiable section and cross-reference topology over one-pass generation on CUAD contracts and EDGAR EX-10 exhibits, and designed BRIEF-L, a dependency-ordered plug-in that lifts within-release coherence from near zero to 0.44–0.70 of cited clauses.
    • Ran retrieval-based source-linkage audits and a cross-generator replication on Llama-3.1-8B; first author of the resulting paper, accepted at the LCFM Workshop, NeurIPS 2026.

    Long-document generation / Graph-conditioned generation / CUAD / EDGAR

  3. March 2026 – May 2026
    LLM Research

    Independent Research Intern

    Shantanu Acharya, Research Scientist at NVIDIA

    • Designed and implemented Pulsar Attention, a distributed long-context LLM inference framework that replaces static anchor duplication with content-aware statistical context summaries.
    • Built the experimental stack for distributed attention, Max-IDF summarization, FlashAttention-based inference, and evaluation on RULER and BABILong.
    • Co-authored the resulting paper, accepted to Findings of EMNLP 2026, showing up to 4.7% accuracy improvement over dense attention, 6.1% over Star Attention, and 3.3x lower Phase-1 FLOPs.

    Distributed attention / FlashAttention / RULER / BABILong

  4. March 2026 – May 2026
    Computer Vision

    Computer Vision Intern

    Auric AI Labs

    • Created an end-to-end synthetic EO data generation pipeline for object-detection models using controlled top-view renders from 3D military-object assets.
    • Developed a GPT-assisted semantic placement pipeline using probabilistic spatial priors to identify contextually valid object locations.
    • Combined BlenderProc rendering, Meshy AI asset generation, and RemoteCLIP-based semantic filtering to improve synthetic scene realism.

    BlenderProc / Meshy AI / RemoteCLIP / Synthetic data