Skip to content
View aryanputta's full-sized avatar
8 followers · 2 following

Achievements

Achievement: Pair ExtraordinaireAchievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pair ExtraordinaireAchievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLO

Highlights

  • Pro

Organizations

@agentrust-io

Block or report aryanputta

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 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
aryanputta/README.md


AI infrastructure · LLM inference · KV cache systems · distributed systems · cloud backends · open-source systems PRs
Rutgers computer science and data science student targeting off-cycle SWE, AI infrastructure, ML systems, cloud backend, and MTS intern roles.
portfolio · merged open-source PRs · LinkedIn

// work

KVCacheForge-X
KV-cache bottleneck lab · TTFT, latency, throughput, HBM stalls, GPU busy, baseline deltas

 RoboFleetOps
AWS-native robotics fleet control plane · Lambda, DynamoDB, SQS, IoT Core, API Gateway, CDK CI
PosCacheBench
Long-context benchmark for positional-attention failure modes under fixed KV-cache budgets

 open-source systems PRs
Reviewed and merged work across NVIDIA, IBM, Microsoft, Hugging Face, FlashAttention, Kubernetes, Pulumi

Availability: off-cycle internships, co-ops, AI infrastructure internships, cloud/backend systems roles, and early MTS-style technical programs.
Current proof standard: every performance claim needs reproducible commands, baseline comparison, hardware/software environment, and an honest limitations section.



// research

satellite telemetry anomaly detection
100K telemetry readings · 5 NASA/ESA fault modes · recurrence-plot CV · 0.91 F1 on Kepler-class wheel oscillation
PDF · repo 		bell labs ml impact analysis
71-paper corpus · semantic clustering · co-authorship networks · Gradient Boosting AUC 0.674 · SHAP attribution
PDF · repo



// open source

31 merged pull requests across NVIDIA · IBM · Microsoft · HuggingFace · LinkedIn · Pulumi · Kubernetes · others — inference, CUDA, CI, and ML-systems internals


NVIDIA/cuda-python#2087
FIPS-safe hashes for program cache keys		 NVIDIA/cuda-quantum#4688
nvqpp: discriminate measured-register bool iteration
huggingface/accelerate#4054
Aggregate profiler memory example		 Dao-AILab/flash-attention#2622
weights_only=True across all torch.load sites
ai-dynamo/dynamo#10281
HTTP 415 for unsupported image formats		 linkedin/Liger-Kernel#1157
Guard save_for_backward on grad_bias in fused linear CE



// stack



// metrics


Pinned Loading

  1. KVCacheX KVCacheX Public

    Memory-aware LLM inference optimizer for KV cache compression, eviction, and scheduling.

    Python

  2. adaptive-compute-runtime adaptive-compute-runtime Public

    Adaptive C++/CUDA runtime that profiles workloads at submission time and dynamically routes to CPU, GPU, or batched execution based on arithmetic intensity and transfer cost

    C++

  3. Helios Helios Public

    Hardware-aware compute runtime in C++ and CUDA for real sparse, dense, and graph workloads.

    C++

  4. IBM/aiu-trace-analyzer IBM/aiu-trace-analyzer Public

    A tool to post-process json trace files for IBM-AIU performance analysis. It enhances the traces with additional statistics extracted from the trace data itself and (optionally) by combining it wit…

    Python 12 7