Skip to content
View amanyagami's full-sized avatar

Block or report amanyagami

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
amanyagami/README.md

Aman Singh

ML Systems Engineer · LLM Training, Serving & AI Infrastructure

I build LLM systems from training and inference to secure, reliable deployment.

Portfolio · Résumé · LinkedIn · Email

Focus: LLM runtimes · GPU/distributed ML · secure execution · robust ML · production systems

Evidence: H100/DDP/FP16-BF16 · FSDP/vLLM · sub-10 ms A6000 inference · Rust sandboxing · 1.8× traffic · 30% lower cost · 60% fewer incidents

Open-source: SGLang · VERL · Megatron-LM · PyTorch AO · LLVM/MLIR · arapuca · ✅ 7 merged

Open-source systems work

Public PR activity

✅ merged — linked counts refresh automatically.

Repository PR status
LeGambiArt/arapuca ✅ 2 merged
MPSLab-ASU/Seperating_OOD_and_ADV ✅ 1 merged
llvm/llvm-project ✅ 4 merged

Auto-updated 2026-09-01 02:14 UTC by update-oss-stats.yml · includes public external PRs authored by amanyagami; excludes personal and excluded repositories

Selected work

Make Presentation Simple · Nandi RAG assistant · ViT benchmarks · DrDNA

Research

Viyog: Separating Adversarial and Out-of-Distribution — accepted at ESWEEK CODES 2026. Research on separating adversarial and OOD inputs using intermediate-representation geometry.

Resources: Venue · Code · PyPI · Leaderboard · Dataset · Weights · Checkpoints

Education: M.S. Computer Engineering, ASU (2024–2026) · B.Tech Electrical and Electronics Engineering, NITK (2018–2022)
Toolkit: Python · C++ · Rust · Go · PyTorch · CUDA · DDP/FSDP · SGLang · vLLM · Kubernetes · Docker · AWS · Google Cloud

Connect

For ML systems, AI infrastructure, or research engineering opportunities: email me or connect on LinkedIn.

Pinned Loading

  1. Fine_Tuning_Vision_Transformers_on_Cifar100 Fine_Tuning_Vision_Transformers_on_Cifar100 Public

    🧠 Fine-tuning Vision Transformers and modern CNNs (ViT, Swin, MobileVit, EfficientNetV2-L ) on CIFAR-100 using PyTorch, timm, and uv.

    Jupyter Notebook 1

  2. Probablistic-modelling-of-features Probablistic-modelling-of-features Public

    This project explores the geometry and probabilistic structure of deep neural network feature spaces, with a focus on class separability, representation collapse, and robustness under adversarial p…

    Jupyter Notebook 1

  3. MPSLab-ASU/Seperating_OOD_and_ADV MPSLab-ASU/Seperating_OOD_and_ADV Public

    A lightweight PyTorch framework for distinguishing out-of-distribution (OOD) inputs from adversarial (ADV) samples using intermediate feature representations.

    Python

  4. SLM-based-QA SLM-based-QA Public

    A Flask-based PDF question answering chatbot that compares direct prompting vs retrieval-augmented generation using Supermemory across multiple small and large language models.

    HTML 1

  5. Detecting-Silent-Data-Corruptions-in-Deep-Neural-Networks Detecting-Silent-Data-Corruptions-in-Deep-Neural-Networks Public

    A PyTorch-based implementation of DrDNA, a post-hoc framework for detecting and mitigating soft errors (SDCs) in deep neural networks. The project profiles layer-wise activation statistics and comp…

    Jupyter Notebook 1

  6. Make_Presentation_Simple.io Make_Presentation_Simple.io Public

    📄➡️📊 Convert PDFs into AI-generated presentation decks using a fully serverless architecture. ⚡ AWS Lambda + Step Functions orchestration 🧠 Multimodal LLM (Qwen) for slide generation ☁️ S3 + Dynamo…

    Python 2 1