The AI Platform Engineer

16 · References & sources

The body of this note is opinionated engineering text; the list below is for verifiable sources on production ML platforms—technical debt, serving, lifecycle tooling, retrieval, agents, and the compliance language execs will use. I use it as a reading list, not as a substitute for your vendor’s contract or your counsel.

Note: Use arXiv / in-press versions for speed; verify the published venue when you need a DOI. Vendor runbooks (Bedrock, Vertex, OpenAI, Anthropic, inference runtimes) change often—treat them as the authority for limits, regions, and pricing.

ML systems, platform practice & lifecycle

1. Sculley et al., “Hidden Technical Debt in Machine Learning Systems.” NeurIPS / NIPS 2015. Foundational essay on why “the model is the easy part”—informs §1–§3 platform boundaries and §13 anti-patterns. NeurIPS proceedings
2. Breck et al., “What’s your ML Test Score? A rubric for ML production readiness.” Google, 2017 (expanded rubric). Checklist DNA behind production gates in §5–§8, §14. Google Research
3. Amershi et al., “Software Engineering for Machine Learning: A Case Study.” ICSE 2019. How mature teams operationalize ML—maps to the shared platform path and role split in §1, §11. https://arxiv.org/abs/1811.04900
4. Polyzotis et al., “Data Management Challenges in Production Machine Learning.” SIGMOD 2017. Data/lineage/registries—underpins §5–§7 and the foundation layer in Figure 1. ACM Digital Library
5. Zaharia et al., “Accelerating the Machine Learning Lifecycle with MLflow.” IEEE Data Eng. Bull. 2018. Experiment tracking, packaging, registry concepts in §5–§6. IEEE Xplore
6. Huyen, Designing Machine Learning Systems. O’Reilly, 2022. End-to-end lifecycle and deployment patterns echoed across §3–§11. ISBN 978-1098107969.
7. Kleppmann, Designing Data-Intensive Applications. O’Reilly, 2017. Storage, streams, and reliability vocabulary for retrieval, logging, and registries (§3–§7). ISBN 978-1449373320.

Kubernetes, scheduling & “cloud-native” serving context

1. Burns et al., “Borg, Omega, and Kubernetes.” ACM Queue / Communications of the ACM 2016. Conceptual background for running inference and control planes on orchestrators (§3, §6, §7 foundation). ACM Queue
2. Verma et al., “Large-scale cluster management at Google with Borg.” EuroSys 2015. Scheduling, quotas, SLO-aware placement—relevant to GPU/tenant fairness in §6, §10. ACM Digital Library

Inference, memory management & throughput

1. Kwon et al., “Efficient Memory Management for Large Language Model Serving with PagedAttention.” SOSP 2023 (vLLM). Core citation for KV-cache/paging discussion in §6. https://arxiv.org/abs/2309.06180
2. Yu et al., “Orca: A Distributed Serving System for Transformer-Based Generative Models.” OSDI 2022. Iteration-level scheduling and batching—adjacent to §6 latency/cost tradeoffs. USENIX OSDI 22
3. Pope et al., “Efficiently Scaling Transformer Inference.” MLSys 2023. Multi-device inference design space. https://arxiv.org/abs/2211.05102
4. Aminabadi et al., “DeepSpeed-Inference: enabling efficient inference of transformer models at unprecedented scale.” SC 2022. High-scale serving stack context. https://arxiv.org/abs/2207.00032

Foundation models, adaptation & training interfaces

1. Vaswani et al., “Attention Is All You Need.” NeurIPS 2017. Transformer substrate for everything in §5–§6. https://arxiv.org/abs/1706.03762
2. Touvron et al., “LLaMA: Open and Efficient Foundation Language Models.” 2023. Open-weights / self-hosting storyline in exec summary, §5–§6. https://arxiv.org/abs/2302.13971
3. Hu et al., “LoRA: Low-Rank Adaptation of Large Language Models.” ICLR 2022. Adapter fine-tuning in Figure 1 layer 6 and §5–§10. https://arxiv.org/abs/2106.09685
4. Rafailov et al., “Direct Preference Optimization (DPO).” NeurIPS 2023. Preference optimization named in §5–§6. https://arxiv.org/abs/2305.18290

RAG, retrieval & hybrid search

1. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.” NeurIPS 2020. Canonical RAG—§7. https://arxiv.org/abs/2005.11401
2. Robertson & Zaragoza, “The Probabilistic Relevance Framework: BM25 and Beyond.” FnTIR 2009. Lexical side of hybrid retrieval in §7. https://dl.acm.org/doi/10.1561/1500000019
3. Karpukhin et al., “Dense Passage Retrieval for Open-Domain Question Answering.” EMNLP 2020. Dense retrieval practice. https://arxiv.org/abs/2004.04906
4. Gao et al., “Retrieval-Augmented Generation: A Survey.” 2024. Single entrypoint for chunking/index/query research. https://arxiv.org/abs/2312.10997

Agents, tools & orchestration protocols

1. Yao et al., “ReAct: Synergizing Reasoning and Acting in Language Models.” ICLR 2023. Agent loop mental model in §2–§3, §7. https://arxiv.org/abs/2210.03629
2. Schick et al., “Toolformer.” 2023. Tool-learning framing. https://arxiv.org/abs/2302.04761
3. Model Context Protocol (MCP). Cross-vendor tool/context standard referenced in §3, §7, §15. https://modelcontextprotocol.io/
4. OpenAPI Specification. HTTP tool contracts at the gateway boundary (§3). https://spec.openapis.org/oas/latest.html

Evaluation, observability & SLO discipline

1. Es et al., “Ragas: Automated Evaluation of Retrieval Augmented Generation.” 2023. RAG quality metrics named in the appendix cheat sheet / §8. https://arxiv.org/abs/2309.15217
2. Zheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena.” NeurIPS 2023. Pairwise judging—§8. https://arxiv.org/abs/2306.05685
3. Beyer et al. (eds.), Site Reliability Engineering. O’Reilly / Google, 2016. SLOs, error budgets, incident practice—§2 cross-cutting, §8, §14. https://sre.google/sre-book/table-of-contents/
4. OpenTelemetry. Vendor-neutral traces/metrics/logs—appendix “OTel everything” and §8. https://opentelemetry.io/
5. W3C Trace Context. Propagation standard for distributed traces through gateways and tool calls. https://www.w3.org/TR/trace-context/

Security, governance & compliance

1. OWASP Top 10 for Large Language Model Applications. Threat taxonomy for §9. OWASP project page
2. Greshake et al., “Indirect Prompt Injection” (real-world LLM-integrated apps). 2023. Pairs with gateway + retrieved-content trust models in §3, §9. https://arxiv.org/abs/2302.12173
3. NIST AI Risk Management Framework (AI RMF 1.0). Risk language for enterprise governance in §9. https://www.nist.gov/itl/ai-risk-management-framework
4. EU Artificial Intelligence Act (Regulation (EU) 2024/1689). Legal primary source for §15 regulatory bullet. EUR-Lex

Identity, multi-tenancy & API security (platform gateway)

1. Hardt et al., “The OAuth 2.0 Authorization Framework.” RFC 6749. Baseline for delegated access patterns at the AI gateway. RFC 6749
2. OpenID Connect Core 1.0. OIDC for AuthN in browser/API flows—referenced conceptually in §3 gateway. OpenID Foundation