Skip to content
View tengli-alaska's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report tengli-alaska

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 user’s behavior. Learn more about reporting abuse.

Report abuse
tengli-alaska/README.md

Alaska Tengli

Systems × AI - from silicon validation to ML infrastructure.

I come from hardware/post-silicon validation, PCIe interconnects, secure boot firmware - and I've been building toward AI infrastructure and performance engineering. I think about inference bottlenecks in terms of cache coherence, memory hierarchies, and what's actually happening on the silicon.

MS in ECE @ Northeastern University · Previously @ Altera (Intel) · Aptiv


Currently

  • Contributing performance fixes to ONNX Runtime - diagnosed and resolved a 17–28% AveragePool regression by identifying a compiler auto-vectorization blocker in the CPU execution provider
  • Research with MIT Sea Grant - performance analysis of CV-based ML models
  • Building MLOps pipelines and distributed systems coursework at Northeastern

Selected Work

AI for Hardware Validation · Altera (Intel FPGA) Built an LLM-driven test plan generator with RAG pipelines and an unsupervised validation engine that improved test coverage by 94%. Designed PCIe/CXL link testing frameworks at the hardware-software interface. Presented at Altera Innovation Day 2025.

GPU & HPC Optimization · Northeastern Heterogeneous CPU–GPU graph workloads (PageRank, Graph Coloring) using MPI + CUDA on NVIDIA A100 clusters. Profiled with Nsight Systems, scheduled with SLURM across multi-node clusters.

Memory Architecture Research · Northeastern Evaluated CXL 2.0 cache coherence protocols (MESI, MOESI, Directory-Based) using QEMU emulation. Presented poster at Future of Memory & Storage 2025.

Embedded Firmware · Aptiv Secure Boot with AES-128 MAC cryptography and HSM integration for automotive ECUs. Debugged FlexRay/SPI paths via Lauterbach TRACE32 and JTAG.


Projects

Project What it does
Workload Clustering Pipeline Airflow DAG for K-Means clustering of server workload telemetry — identifies CPU-bound, memory-bound, GPU-intensive, and idle profiles
ML Inference Latency Predictor Estimates inference time and throughput (FPS) across cloud GPUs, edge devices, and CPUs — deployed on GCP Cloud Run
BlueForecast MLOps Pipeline End-to-end demand prediction with XGBoost, Airflow, DVC, and GCP across 8M+ rows
ONNX Runtime (fork) Fix for AveragePool performance regression — guarded bounds clamping behind ceil_mode check to restore auto-vectorization
Distributed Systems Cloud-native services in Go — MapReduce, containerized APIs on AWS ECS Fargate, Terraform IaC

Tech

Systems       C · C++ · Verilog · PCIe/CXL · RISC-V · ARM · FreeRTOS · QEMU · JTAG · GDB · Valgrind
AI/ML         Python · PyTorch · TensorFlow · ONNX Runtime · CUDA · Nsight Systems · Hugging Face · LangChain
Infra         Go · Docker · Terraform · AWS · GCP · Jenkins · Airflow · DVC · gRPC
HPC           MPI · OpenMP · SLURM · CUDA Profiling · Latency/Bandwidth Analysis

Publications & Talks


LinkedIn Email

Pinned Loading

  1. workload-clustering-airflow workload-clustering-airflow Public

    Apache Airflow DAG pipeline for K-Means clustering of server workload telemetry data. Identifies workload profiles (CPU-bound, memory-bound, GPU-intensive, idle) using the elbow method. Runs in Doc…

    Python

  2. infra-starter-repo infra-starter-repo Public

    Forked from RuidiH/CS6650_2b_demo

    HCL

  3. Distributed-Systems Distributed-Systems Public

    Go

  4. onnxruntime onnxruntime Public

    Forked from microsoft/onnxruntime

    ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator

    C++

  5. BlueForecast--Predictive-Operations-Platform-for-Bluebikes BlueForecast--Predictive-Operations-Platform-for-Bluebikes Public

    Jupyter Notebook