Indian MLOps and LLMOps Group (IMLG)

Objective

IMLG exists to build deep, production-grade fluency in the infrastructure that runs modern AI — from the silicon up. Every session is part of a structured curriculum: GPU architecture, the CUDA and library stack, distributed training frameworks, MLOps and LLMOps pipelines, observability, and deployment across edge, server, and cloud. The goal is not survey knowledge — it is the kind of understanding that lets you reason about bottlenecks, make architectural decisions, and operate AI systems at scale.

Beyond the technical curriculum, 𝐈𝐌𝐋𝐆 𝐢𝐬 𝐜𝐨𝐦𝐦𝐢𝐭𝐭𝐞𝐝 𝐭𝐨 𝐧𝐮𝐫𝐭𝐮𝐫𝐢𝐧𝐠 𝐚 𝐦𝐚𝐭𝐮𝐫𝐞 𝐌𝐋𝐎𝐩𝐬 𝐚𝐧𝐝 𝐋𝐋𝐌𝐎𝐩𝐬 𝐩𝐫𝐚𝐜𝐭𝐢𝐜𝐞 𝐢𝐧 𝐈𝐧𝐝𝐢𝐚 — 𝐜𝐨𝐧𝐧𝐞𝐜𝐭𝐢𝐧𝐠 𝐞𝐧𝐭𝐡𝐮𝐬𝐢𝐚𝐬𝐭𝐬, 𝐩𝐫𝐚𝐜𝐭𝐢𝐭𝐢𝐨𝐧𝐞𝐫𝐬, 𝐢𝐧𝐝𝐮𝐬𝐭𝐫𝐲 𝐞𝐱𝐩𝐞𝐫𝐭𝐬, 𝐭𝐡𝐞 𝐬𝐭𝐚𝐫𝐭𝐮𝐩 𝐞𝐜𝐨𝐬𝐲𝐬𝐭𝐞𝐦, 𝐚𝐧𝐝 𝐚𝐜𝐚𝐝𝐞𝐦𝐢𝐚 𝐢𝐧𝐭𝐨 𝐚 𝐬𝐢𝐧𝐠𝐥𝐞 𝐜𝐨𝐦𝐦𝐮𝐧𝐢𝐭𝐲 𝐭𝐡𝐚𝐭 𝐫𝐚𝐢𝐬𝐞𝐬 𝐭𝐡𝐞 𝐜𝐨𝐥𝐥𝐞𝐜𝐭𝐢𝐯𝐞 𝐛𝐚𝐫.

Group Description

Most AI communities stop at model APIs and framework tutorials. IMLG goes further.

We are a practitioner group for engineers, architects, and platform builders who want to understand how AI infrastructure actually works — not just use it. Our sessions follow a progressive curriculum that begins with GPU compute fundamentals and builds through the CUDA software stack, distributed training strategies, MLOps and LLMOps lifecycle management, and production observability — before examining how all of it maps onto IoT, edge, on-prem, and cloud environments.

Topics we go deep on:

• GPU architecture: streaming multiprocessors, Tensor Cores, HBM, NVLink and NVSwitch topologies
• CUDA programming model, Triton, and the NVIDIA library ecosystem — cuBLAS, NCCL, NIXL, cuDNN
• Intra-node and inter-node communication: PCIe, GPUDirect, RDMA, DPU
• Distributed training: data, model, pipeline, and tensor parallelism; FSDP; Megatron-LM; torch.compile
• MLOps: experiment tracking, feature stores, vector and graph databases, serving pipelines
• LLMOps: KV cache management, speculative decoding, PEFT, RLHF strategies, inference optimization
• GPU and training observability: DCGM, Nsight, OpenTelemetry, Prometheus, ClickHouse, Grafana
• LLM inference observability: OpenLLMetry, Langfuse, Phoenix, TTFT and ITL SLOs
• Platform Engineering and Linux administration for GPU infrastructure
• Deployment across edge (Jetson, NPU), on-prem GPU servers, and cloud GPU instances

This is a Hyderabad-based group with a national reach. Sessions are technical, hands-on where possible, and sequenced — each meetup builds on the last. If you are building or operating AI systems in production and want to go below the abstraction layer, this is your community.