Saurabh Ghatnekar is an AI engineer working across the LLM stack: training and inference systems (kernels, parallelism, quantization), data pipelines (curation, deduplication, mixing), and alignment (SFT, RL, preference and synthetic data). Deep-dive writing on how frontier models actually get built.https://saurabh.works/en-ushttps://saurabh.works/blog/sft-dpo-rlhf-post-training-guide/https://saurabh.works/blog/sft-dpo-rlhf-post-training-guide/When supervised fine-tuning is enough, when preference optimization pays off, and where verifiable rewards fit — a practical decision guide.Fri, 03 Jul 2026 00:00:00 GMTalignmentpost-trainingsftdporlhfpreference-dataverifiersconnect@saurabh.works (Saurabh Ghatnekar)https://saurabh.works/blog/llm-data-curation-filtering-deduplication-mixing/https://saurabh.works/blog/llm-data-curation-filtering-deduplication-mixing/A practical walkthrough of the LLM data pipeline — quality filtering, exact and near deduplication with MinHash, decontamination, and mixture weights.Thu, 25 Jun 2026 00:00:00 GMTdatadata-curationdeduplicationminhashdata-mixingllm-pretrainingconnect@saurabh.works (Saurabh Ghatnekar)https://saurabh.works/blog/fit-large-language-models-on-small-gpus/https://saurabh.works/blog/fit-large-language-models-on-small-gpus/Where GPU memory actually goes during LLM training, and how activation checkpointing, quantization, 8-bit optimizers, and CPU offloading win it back.Fri, 12 Jun 2026 00:00:00 GMTsystemsgpu-memoryactivation-checkpointingquantizationcpu-offloadingllm-trainingconnect@saurabh.works (Saurabh Ghatnekar)