mohsinm-dev - Overview

Mohsin Mehmood

Machine Learning Engineer · LLM Systems & Agent Architectures · Open-Source Contributor

Forward Deployed ML Engineer at Kodamai (Glasgow). I design and build production core ML and AI systems: multi-agent pipelines, document extraction infrastructure, and LLM-powered workflow automation. Most of my work sits at the boundary between production-grade ML and the messy reality of deploying it in regulated, high-stakes environments.

I carefully consider system architecture before writing code, including transaction boundaries, failure modes, inference latency budgets, and how models degrade in production. I have shipped AI systems in healthcare (HIPAA/PDPA), enterprise supply chain, and early-stage startups, working directly with non-technical domain experts to translate complex processes into automated pipelines.


Open-Source Contributions

Project Contribution
google/flax #5 contributor (last 12 months, official leaderboard · 19 commits). Merged PRs in Flax NNX core: fixed nnx.tabulate crash with empty/None values (#4891); fixed variable hook display bugs in nnx.tabulate (#5008). Both shipped in v0.12.1.
python/cpython 6 merged PRs: fixed TextIOWrapper.tell() assertion failure with standalone carriage return (backported to 3.13 + 3.14 · #141331); fixed re.Match.group() doc claiming [1..99] range limit (#144696); documented asyncio Task cancellation propagation (#141249); fixed inaccurate object comparison docs (#141221).
pytorch/pytorch Approved PR in PyTorch core (#167209).
uber/causalml Fixed bug where estimation_sample_size was not propagated from UpliftRandomForestClassifier to individual trees (#850). Silently broken behaviour in the core uplift Cython layer.

Technical Depth

core_ml = {
    "frameworks":     ["PyTorch", "JAX/Flax", "TensorFlow"],
    "training":       ["LoRA", "QLoRA", "RLHF", "DPO", "mixed precision", "distributed training"],
    "inference":      ["quantization (GPTQ/AWQ/bitsandbytes)", "vLLM", "GGUF/MLX", "TensorRT"],
    "evaluation":     ["RAGAS", "ROUGE", "BERTScore", "bootstrapped CI", "MLflow", "W&B"],
    "architectures":  ["Transformers", "CNNs", "multi-agent systems", "RAG pipelines"],
}

systems = {
    "design":         ["modular monolith", "worker architecture", "state machines", "transactional outbox"],
    "infra":          ["Docker", "Kubernetes", "FastAPI", "Redis", "RabbitMQ", "Celery"],
    "cloud":          ["GCP Vertex AI", "AWS SageMaker / EC2 / Lambda", "Azure ML"],
    "observability":  ["structured logging", "latency tracing", "drift monitoring", "PHI/PII redaction"],
    "compliance":     ["HIPAA", "Singapore PDPA"],
    "languages":      ["Python", "C++", "Bash"],
}

Contact