Hayvin
Hayvin
How It WorksData TypesUse CasesBrainstormSample Profile
Hayvin
Hayvin

Professional networking based on deep interests, not addictive reinforcement.

Product

  • Data Types
  • Use Cases
  • Brainstorm Agent
  • Sample Profile
  • Hackathon

Company

  • About
  • Blog
  • Careers
  • Contact

Legal

  • Privacy Policy
  • Terms of Service
  • Data Usage

© 2026 Hayvin. San Francisco • Privacy-first networking

GraphRAG powered
SAMPLE PROFILE · MOCK DATA
AC

Alex Chen

Graph score: 0.91

Inference Engineer · Scale AI

San Francisco, CAachen-ml · 847 contributions4 arXiv papers · 2 ICLR reviewsOpen for coffee · Tue/Thu

Powered by

GitHub12 nodes
arXiv4 nodes
OpenReview2 nodes
Résumé6 nodes

Expertise scores

computed · not self-reported

CUDA / GPU Kernels

GitHub

0.91

Sparse Attention

arXiv + GitHub

0.88

KV Cache Optimization

GitHub + OpenReview

0.85

Distributed Training

GitHub

0.79

Quantization (FP8/INT4)

Papers

0.76

Inference Serving

GitHub

0.73

RLHF / Fine-tuning

OpenReview

0.61

Triton Kernels

GitHub

0.58

Recent activity

vLLMMay 2025

3 merged PRs — PagedAttention perf, chunked prefill memory

arXiv:2504.11823Apr 2025

Efficient KV Cache Eviction via Graph-Structured Importance Scoring

flash-attnMar 2025

Issue #812 — sliding window + GQA correctness fix

ICLR 2025Feb 2025

2 papers reviewed — sparse MoE routing, long-context eval

arXiv:2502.07341Feb 2025

Speculative Decoding with Draft Token Pruning

TensorRT-LLMJan 2025

Custom attention plugin for GQA + sliding window

Agent recommendations

Engineers most likely to collaborate with Alex based on graph proximity, shared contributions, and topic overlap.

AT
Alex T.0.94

Research Eng · DeepMind · 1.2mi

3 co-cited papers on KV eviction; both contributed to vLLM in the last 90 days.

KV CachePagedAttnFlashAttn
YL
Yun L.0.89

PhD · Stanford NLP · 4.1mi

Sparse attention co-author; working on similar long-context efficiency problems for LLM serving.

Sparse AttnGQALong Context
MK
Min K.0.86

MLSys · Together AI · 2.8mi

Active on flash-attn and vLLM; similar CUDA kernel optimization background.

TritonCUDABatching

Powered by graph traversal + embedding similarity

Score breakdown

Contribution depth
0.93
Recency
0.88
Peer validation
0.86
Topic consistency
0.79
Network proximity
0.71

See your own profile.

Join the beta, connect your data sources, and see how Hayvin scores and surfaces your expertise.