Skip to content
View aasthavar's full-sized avatar
๐ŸŽฏ
Focusing
๐ŸŽฏ
Focusing

Block or report aasthavar

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
aasthavar/README.md

๐Ÿ‘‹ Hi, Iโ€™m Aastha

Applied ML researcher/engineer. I build scalable LLM systems, agentic AI, and robotics, bridging research โ†’ production.


โšก Highlights

  • LLM Pre-training & Fine-tuning: Distributed pipelines (DeepSpeed, FSDP, SMP, LoRA/PEFT, CPU offload) for 7Bโ€“70B models โ†’ training 57h โ†’ 5.6h, multi-GPU/Node, benchmarked and scaled for production.

  • Inference Optimization: Triton + TensorRT + vLLM, fused attention, multi-LoRA adapters, quantization โ†’ 70% latency reduction, 80+ QPS, production-grade deployments.

  • Embodied & Agentic AI: Modular exoskeleton (LLM + vision + speech) โ†’ ICML 2025 demo. Multi-agent orchestration for GenAI campaigns & designer assistant with RAG, query rewriting, hallucination detection.

  • Mechanistic Interpretability: Crosscoders (sparse autoencoders) to probe LLM instruction-tuning; HF open-source pipeline.

  • NP-hard / HPC Projects: Brick Maestro: Lego assembly optimization using HPC, AWS ParallelCluster โ†’ presented at AWS re:Invent, Paris Summit.

  • Foundations: Deep Learning for Face Anti-Spoofing (Thesis), TA at NIT Rourkela, Algorithm (karatsuba + quad itoh-tsuji) optimization @ DRDO.


๐Ÿ› ๏ธ Tech Stack

  • Training: PyTorch, DeepSpeed, FSDP, SMP, LoRA/PEFT, Multi-GPU/Node
  • Inference: TensorRT-LLM, vLLM, sglang, LoRA adapters, Quantization & Distillation
  • Infra: AWS (SageMaker, EKS, HyperPod, ParallelCluster), Docker, Prometheus, Grafana
  • Other: HPC, distributed LLM scaling, agentic AI

Pinned Loading

  1. inference-bench inference-bench Public

    Python

  2. grok-ml-dl grok-ml-dl Public

    A personal worklog as I learn and implement the core concepts behind machine/deep learning

    Jupyter Notebook

  3. crosscoders-qwen crosscoders-qwen Public

    Jupyter Notebook

  4. synthetic-ner-data-train-comprehend synthetic-ner-data-train-comprehend Public

    Train NER model with PDF annotations. Entites is generated using llms and annotations are constructed with textract blocks.

    Jupyter Notebook 1

  5. talk-to-your-logs talk-to-your-logs Public

    Find the root cause of issues using logs of an application.

    Jupyter Notebook

  6. finetune-evaluate-codestral finetune-evaluate-codestral Public

    Different approaches for finetuning, evaluating, optimizations for code generation model - codestral

    Jupyter Notebook 11