ONLINE suhas@tübingen
CEST
Operator ml infrastructure

Suhas Goravale Siddaramu

I work on the layer between a model and the metal it runs on.

Five years of getting models to run somewhere real — distributed across a Slurm cluster, served through a traced inference pipeline, or running entirely offline on someone's own machine.

Role
Computer Research Scientist, AI
Site
Universitätsklinikum Tübingen
Location
Reutlingen, Germany
Focus
Distributed training on Slurm · LLM PEFT fine-tuning · Quantization & inference optimization · Agentic RAG
What I work on model → metal → served
Train Serve Model · Data PyTorch · TF HuggingFace Autotrainer distributed training LoRA · LoReFT PEFT · 4-bit Slurm · DDP · FSDP GPU · Multi-Node SlurmGenie cluster copilot Quantize · Prune vLLM · ONNX · GGUF RAG · Agents FAISS · Ollama · LangChain InfraSight traces · evals Served
Job history sacct · one shared axis
    Job queue squeue · state from last push
      RUNNING — pushed in the last 60 days IDLE — within the year COMPLETED — older
      Archive earlier repositories
        Stack what I reach for

        LLM & retrieval

        • HuggingFace
        • PEFT
        • LoRA
        • LoReFT
        • BitsAndBytes
        • LangChain
        • Ollama
        • FAISS
        • Vector DBs

        Training & inference

        • PyTorch
        • DDP
        • FSDP
        • TensorFlow
        • Keras
        • ONNX
        • TVM
        • scikit-learn
        • OpenCV

        Clusters & MLOps

        • Slurm
        • Docker
        • Kubernetes
        • Podman
        • Argo Workflows
        • GitLab CI
        • GitHub Actions
        • GCP
        • AWS

        Programming

        • Python
        • Bash
        • SQL
        • C

        Data

        • pandas
        • NumPy
        • Matplotlib
        • Seaborn

        Environment

        • Linux
        • uv
        • Anaconda
        • LaTeX
        • Jira
        Education & languages
        • M.Sc. Embedded Systems Technische Universität Chemnitz 2017–2021
        • B.E. Electronics & Communication Visvesvaraya Technological University, India 2012–2016
        • EnglishNative
        • KannadaNative
        • GermanB1
        Contact

        Open for new opportunities.

        Happy to talk about LLM fine-tuning, agentic pipelines, distributed training, cluster scheduling, or getting a model small enough to actually ship.