Skip to content
View KaranSinghDev's full-sized avatar
6 followers · 5 following

Achievements

Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: QuickdrawAchievement: Pull Shark

Achievements

Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: QuickdrawAchievement: Pull Shark

Highlights

  • Pro

Block or report KaranSinghDev

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
KaranSinghDev/README.md

Karan Singh

ORCID GitHub Email

I am a CS engineer working in Machine Learning and High-Performance Systems, with a background in building scientific computing tools for CERN-HSF (GSoC '25). My focus is on low-latency infrastructure, distributed systems, and empirical AI evaluation.

🧠 Machine Learning & AI Evaluation

Project Problem $\rightarrow$ Solution Core Tech
Valence-Lens Asking models how they feel is unreliable $\rightarrow$ Uses residual-stream activation probes to read internal states directly. PyTorch Interpretability
Loyalty-Lens Models can hide covert allegiances that black-box chats miss $\rightarrow$ Uses white-box probes to detect weight-installed loyalties. PyTorch DOI
StaleBench RAG pipelines silently return outdated answers when facts change $\rightarrow$ A deterministic benchmark measuring answer freshness. Python DOI
DisElect-Africa Safety filters are rarely tested on non-Western contexts $\rightarrow$ Audits election disinformation and safety-prompting trade-offs. Red-Teaming Evals

⚡ Systems & High-Performance Computing

Project Problem $\rightarrow$ Solution Core Tech
RNTuple-Arrow-Gateway Converting physics data between formats wastes memory $\rightarrow$ A C++ gateway enabling zero-copy data flow to Apache Arrow. C++17 Apache Arrow ROOT
AXON-HLT Real-time AI decisions choke single machines $\rightarrow$ Offloads compute to NVIDIA Triton via dynamic batching under a 7ms budget. C++ NVIDIA Triton
E-QUEST Accuracy metrics ignore energy costs $\rightarrow$ A framework measuring physical power and memory for classical vs. quantum ML. Quantum ML PyTorch
WebROOT_X Server-side physics analysis doesn't scale sustainably $\rightarrow$ Runs C++ physics kernels directly in the browser via WebAssembly. C++17 WebAssembly

🌐 Distributed Systems & Data Infrastructure

Project Problem $\rightarrow$ Solution Core Tech
Graviton Scheduler Standard Kubernetes schedules compute blindly across networks $\rightarrow$ A Go Operator that places jobs where the data physically lives. Go Kubernetes
Insight-TSDB Telemetry requires fast storage and live health checks $\rightarrow$ A custom C++17 TSDB paired with a real-time Spark anomaly pipeline. C++17 Kafka Spark
Distributed-Ops-Hub Databases crash when nodes fail $\rightarrow$ A leaderless key-value store with consistent hashing and automated chaos-test resilience. Python gRPC Chaos Testing
Arkhe Archive Research archives are too heavy to run locally $\rightarrow$ A lightweight, single-command data archive adhering to the FAIR standard. FastAPI PostgreSQL DOI

🛠️ Technical Skills

  • Core Languages: C++, Python, Go, Bash
  • Machine Learning & Evals: PyTorch, NVIDIA Triton/CUDA, Mechanistic Interpretability, Benchmarking
  • Systems & Cloud: Kubernetes, Apache Kafka, Apache Spark, gRPC, Docker, PostgreSQL
  • Reliability & Rigor: Valgrind (Memory Profiling), GoogleTest, CMake, CI/CD, Chaos Engineering

Pinned Loading

  1. Arkhe-Open-Data-Archive Arkhe-Open-Data-Archive Public

    A self-hostable scientific data repository for research groups and academic environments. Based on CERN ZENODO

    Python 1

  2. AXON-HLT AXON-HLT Public

    A reference implementation of the SONIC architectural pattern for decoupled GPU inference in HEP trigger systems.

    Python

  3. Graviton-Data-Local-Scheduler Graviton-Data-Local-Scheduler Public

    Kubernetes operator that eliminates WAN data transfers by scheduling compute co-located with its storage replicas, built for distributed physics data lakes.

    Go

  4. RNTuple-Arrow-Gateway RNTuple-Arrow-Gateway Public

    A sophisticated bridge between RNTuple and Apache Arrow for High Luminosity LHC data workflows

    C++

  5. Trace-Initiative/Loyalty-Lens Trace-Initiative/Loyalty-Lens Public

    Detecting hidden loyalties in language models by reading their activations, not their answers.

    Python

  6. Trace-Initiative/Valence-Lens Trace-Initiative/Valence-Lens Public

    A tool that reads the LLM's internal state directly from its weights, tracking internal signals when output text fails.

    Python