6. Production

Infrastructure, Safety, and Capstones

1. Managed LLM Platforms — Bedrock, Azure OpenAI, Vertex AI
2. Inference Platform Economics — Fireworks, Together, Baseten, Modal
3. GPU Autoscaling on Kubernetes — Karpenter, KAI Scheduler
4. Serving Engine Internals — PagedAttention, Continuous Batching, Chunked Prefill
5. EAGLE-3 Speculative Decoding in Production
6. Prefix-Cache Serving — RadixAttention and KV Reuse
7. Hardware-Specialized Inference Compilation — FP8 and NVFP4 on Blackwell
8. Inference Metrics — TTFT, TPOT, ITL, Goodput, P99
9. Production Quantization — AWQ, GPTQ, GGUF, FP8, NVFP4
10. Cold Start Mitigation for Serverless LLMs
11. Multi-Region LLM Serving and KV Cache Locality
12. Edge Inference — ANE, Hexagon, WebGPU, Jetson
13. LLM Observability Stack Selection
14. Prompt Caching and Semantic Caching Economics
15. Batch APIs — the 50% Discount as Industry Standard
16. Model Routing as a Cost-Reduction Primitive
17. Disaggregated Prefill/Decode — NVIDIA Dynamo and llm-d
18. Production Serving Stack — KV Offloading and Cache-Aware Routing
19. AI Gateways — LiteLLM, Portkey, Kong, Bifrost
20. Shadow, Canary, and Progressive Deployment
21. A/B Testing LLM Features — GrowthBook and Statsig
22. Load Testing LLM APIs — k6, LLMPerf, GenAI-Perf
23. SRE for AI — Multi-Agent Incident Response
24. Chaos Engineering for LLM Production
25. Security — Secrets, PII Scrubbing, Audit Logs
26. Compliance — SOC 2, HIPAA, GDPR, EU AI Act, ISO 42001
27. FinOps for LLMs — Unit Economics and Multi-Tenant Attribution
28. Self-Hosted Serving Selection — Matching Engine to Hardware and Scale
29. Model Demo Interfaces — From a Function to a Shareable App
30. Instruction-Following as Alignment Signal
31. Reward Hacking & Goodhart's Law
32. Direct Preference Optimization Family
33. Sycophancy as RLHF Amplification
34. Constitutional AI & RLAIF
35. Mesa-Optimization & Deceptive Alignment
36. Sleeper Agents — Persistent Deception
37. In-Context Scheming in Frontier Models
38. Alignment Faking
39. AI Control — Safety Despite Subversion
40. Scalable Oversight & Weak-to-Strong
41. Red-Teaming: PAIR & Automated Attacks
42. Many-Shot Jailbreaking
43. ASCII Art & Visual Jailbreaks
44. Indirect Prompt Injection
45. Red-Team Tooling: Garak, Llama Guard, PyRIT
46. WMDP & Dual-Use Capability Evaluation
47. Frontier Safety Frameworks — RSP, PF, FSF
48. Model Welfare Research
49. Bias & Representational Harm
50. Fairness Criteria: Group, Individual, Counterfactual
51. Differential Privacy for LLMs
52. Watermarking: SynthID, Stable Signature, C2PA
53. Regulatory Frameworks: EU, US, UK, Korea
54. EchoLeak & CVEs for AI
55. Model, System & Dataset Cards
56. Data Provenance & Training-Data Governance
57. Alignment Research Ecosystem: MATS, Redwood, Apollo, METR
58. Moderation Systems: OpenAI, Perspective, Llama Guard
59. Dual-Use Risk: Cyber, Bio, Chem, Nuclear
60. Terminal-Native Coding Agent
61. RAG over Codebase (Cross-Repo Semantic Search)
62. Real-Time Voice Assistant (ASR → LLM → TTS)
63. Multimodal Document QA (Vision-First)
64. Autonomous Research Agent (AI-Scientist Class)
65. DevOps Troubleshooting Agent for Kubernetes
66. End-to-End Fine-Tuning Pipeline
67. Production RAG Chatbot (Regulated Vertical)
68. Code Migration Agent (Repo-Level Upgrade)
69. Multi-Agent Software Engineering Team
70. LLM Observability & Eval Dashboard
71. Video Understanding Pipeline (Scene → QA)
72. Stateless MCP Server with Registry and Governance
73. Speculative-Decoding Inference Server
74. Constitutional Safety Harness + Red-Team Range
75. GitHub Issue-to-PR Autonomous Agent
76. Personal AI Tutor (Adaptive, Multimodal)
77. Agent Harness Loop Contract
78. Tool Registry with Schema Validation
79. JSON-RPC 2.0 Over Newline-Delimited Stdio
80. Function Call Dispatcher
81. Plan-Execute Control Flow
82. Verification Gates and Observation Budget
83. Sandbox Runner with Denylist and Path Jail
84. Eval Harness with Fixture Tasks
85. Observability with OTel GenAI Spans and Prometheus Metrics
86. End-to-End Coding Agent on the Harness
87. BPE Tokenizer From Scratch
88. Tokenized Dataset with Sliding Window
89. Token and Positional Embeddings
90. Multi-Head Self-Attention
91. Transformer Block from Scratch
92. GPT Model Assembly
93. Training Loop and Evaluation
94. Loading Pretrained Weights
95. Classifier Fine-Tuning by Head Swap
96. Instruction Tuning by Supervised Fine-Tuning
97. Direct Preference Optimization from Scratch
98. Full Evaluation Pipeline
99. Large Corpus Downloader
100. HDF5 Tokenized Corpus
101. Cosine LR with Linear Warmup
102. Gradient Clipping and Mixed Precision
103. Gradient Accumulation
104. Checkpoint Save and Resume
105. Distributed Data Parallel and FSDP from Scratch
106. Language Model Evaluation Harness
107. Hypothesis Generator
108. Literature Retrieval
109. Experiment Runner
110. Result Evaluator
111. Paper Writer
112. Critic Loop
113. Iteration Scheduler
114. End-to-End Research Demo
115. Vision Encoder Patches
116. Vision Transformer Encoder
117. Projection Layer for Modality Alignment
118. Cross-Attention Fusion
119. Vision-Language Pretraining
120. Multimodal Evaluation
121. Chunking Strategies, Compared
122. Hybrid Retrieval with BM25 and Dense Embeddings
123. Cross-Encoder Reranker
124. Query Rewriting: HyDE, Multi-Query, and Decomposition
125. RAG Evaluation: Precision, Recall, MRR, nDCG, Faithfulness, Answer Relevance
126. End-to-End RAG System
127. Task Spec Format
128. Classical Metrics
129. Code Exec Metric
130. Perplexity and Calibration
131. Leaderboard Aggregation
132. End-to-End Eval Runner
133. Collective Ops From Scratch
134. Data Parallel DDP From Scratch
135. ZeRO Optimizer State Sharding
136. Pipeline Parallel and Bubble Analysis
137. Sharded Checkpoint and Atomic Resume
138. End-to-End Distributed Training
139. Jailbreak Taxonomy
140. Prompt Injection Detector
141. Refusal Evaluation
142. Content Classifier Integration
143. Constitutional Rules Engine
144. End-to-End Safety Gate