MS in Large Language Model Engineering

Large language model neural network visualization

Program Overview

The MS in Large Language Model Engineering is the flagship program of the School of Language & Reasoning. Students gain deep, hands-on expertise in the full LLM lifecycle: from tokenization and architecture design through pretraining, fine-tuning, alignment, and production deployment. Graduates leave Meridian AI equipped to build and ship language model systems at scale.

What You Will Learn

The curriculum is organized around the LLM engineering stack:

  • Foundations: Tokenization (BPE, WordPiece, SentencePiece), transformer architectures (GPT, BERT, T5 families), attention mechanisms, positional encoding
  • Training: Dataset curation, pretraining at scale, distributed training (FSDP, DeepSpeed), mixed precision, gradient checkpointing
  • Alignment: RLHF, DPO, Constitutional AI, reward modeling, red-teaming
  • Inference: Quantization (GPTQ, AWQ, GGUF), speculative decoding, KV-cache optimization, serving with vLLM and TGI
  • Applications: RAG systems, function calling, multi-agent architectures, MCP integration

Feste Curriculum Thread

A distinctive feature of this program is the Feste thread: a semester-long implementation project using Tag1 Consulting's open-source Rust LLM engine as the primary teaching vehicle. Students work through the Feste blog series (Parts 1–4, published at tag1.com/how-to/) and extend it with their own modifications. This hands-on approach — reading actual production code, not toy examples — is one of the things Meridian AI graduates say most distinguishes their education.

Career Outcomes

Graduates enter roles as ML engineers, LLM platform engineers, applied research scientists, and AI infrastructure leads at technology companies, AI labs, and enterprise organizations. Median starting salary for the MS class of 2025: $178,000.

Sample Courses

  • LLM-401: Transformer Architecture and Attention Mechanisms
  • LLM-402: Tokenization, Vocabulary, and Encoding Schemes
  • LLM-410: Large-Scale Pretraining: Data, Compute, and Infrastructure
  • LLM-420: Fine-Tuning and Parameter-Efficient Adaptation (LoRA, QLoRA)
  • LLM-430: Alignment: RLHF, DPO, and Constitutional AI
  • LLM-440: Inference Optimization and Deployment
  • LLM-450: Agentic Systems and Multi-Agent Orchestration
  • LLM-490: Capstone: Build and Ship an LLM Application