# 12-Month Advanced AI Infrastructure > The long arc. A year of compounding depth across search, vector databases, AI systems, backend, storage, and system design — measured in shipped artifacts and project features. Canonical: https://learn.significanthobbies.com/curriculum/roadmaps/ai-infra-12-month The long arc. A year of compounding depth across search, vector databases, AI systems, backend, storage, and system design — measured in shipped artifacts and project features. - Horizon: 12mo - Outcome: Reach the depth to design and reason about systems like Turbopuffer, and ship them through HighSignal and Codevetter. - Tracks: Search & IR, Vector DB & ANN, AI Systems, Backend, Databases & Storage, System Design Work each milestone as a claim that needs evidence. Predict the mechanism and failure mode first, study the linked concepts, complete the drills without copying an answer, and preserve the resulting code, measurements, or design. ## Milestone 1: Prerequisites — unlock the rest of this path Concepts the later milestones depend on. Without these the planner cannot serve the rest of this roadmap. ### Concepts - [ML Math Foundations](https://learn.significanthobbies.com/curriculum/concepts/ml-math) — Vectors, matrices, dot products, matmul, shapes. - [Probability Fundamentals](https://learn.significanthobbies.com/curriculum/concepts/probability-fundamentals) — Sample spaces, conditional probability, independence, and Bayes' rule for updating beliefs. - [Descriptive Statistics](https://learn.significanthobbies.com/curriculum/concepts/descriptive-statistics) — Summarizing data: mean, median, variance, correlation, and when each summary lies. - [Numerical Stability](https://learn.significanthobbies.com/curriculum/concepts/numerical-stability) — Floating point, catastrophic cancellation, and the log-sum-exp trick. - [Capacity Estimation](https://learn.significanthobbies.com/curriculum/concepts/capacity-estimation) — Back-of-the-envelope QPS, storage, and bandwidth maths that sizes a design before you draw it. - [Requirements Scoping](https://learn.significanthobbies.com/curriculum/concepts/requirements-scoping) — Separating functional from non-functional requirements and cutting scope to something buildable. - [Embeddings](https://learn.significanthobbies.com/curriculum/concepts/embeddings) — Mapping text/images into dense vectors where distance encodes meaning. - [HTTP Lifecycle](https://learn.significanthobbies.com/curriculum/concepts/http-lifecycle) — DNS → TCP/TLS → request → response: status codes, headers, keep-alive. - [Structured Outputs](https://learn.significanthobbies.com/curriculum/concepts/structured-outputs) — Forcing LLM output into a validated JSON schema. - [Tokenization](https://learn.significanthobbies.com/curriculum/concepts/tokenization) — Splitting text into terms: lowercasing, stemming, stop words, n-grams. - [Tool Calling](https://learn.significanthobbies.com/curriculum/concepts/tool-calling) — Letting an LLM invoke functions/APIs via structured calls. - [Vector Similarity](https://learn.significanthobbies.com/curriculum/concepts/vector-similarity) — Cosine, dot product, and L2 distance — how to score vector closeness. - [Inverted Index](https://learn.significanthobbies.com/curriculum/concepts/inverted-index) — Term → posting list mapping that powers fast keyword lookup. - [TF-IDF](https://learn.significanthobbies.com/curriculum/concepts/tf-idf) — Term frequency × inverse document frequency weighting for relevance. - [Top-k Vector Search](https://learn.significanthobbies.com/curriculum/concepts/topk-vector-search) — Returning the k nearest vectors to a query, exact or approximate. - [Brute-Force Vector DB](https://learn.significanthobbies.com/curriculum/concepts/brute-force-vector-db) — Exact nearest-neighbour search by scanning every vector — the correctness baseline. ## Milestone 2: Q1 — Retrieval and vector foundations Complete the 90-day AI search/infra roadmap. ### Concepts - [BM25](https://learn.significanthobbies.com/curriculum/concepts/bm25) — The standard lexical ranking function: TF saturation + IDF + length normalization. - [Hybrid Search](https://learn.significanthobbies.com/curriculum/concepts/hybrid-search) — Fusing lexical (BM25) and vector retrieval, usually via reciprocal rank fusion. - [HNSW](https://learn.significanthobbies.com/curriculum/concepts/hnsw) — Hierarchical Navigable Small World graphs — the dominant ANN index. - [Metadata Filtering](https://learn.significanthobbies.com/curriculum/concepts/metadata-filtering) — Combining vector search with structured predicates (tenant, date, tags). - [RAG](https://learn.significanthobbies.com/curriculum/concepts/rag) — Retrieval-Augmented Generation: ground an LLM answer in retrieved context. ### Build evidence - **Hybrid search v0 in HighSignal** — Combine BM25 and vector retrieval with reciprocal rank fusion. - **Brute-force vector index** — An exact nearest-neighbour index — the recall=1.0 baseline. - **RAG pipeline v0** — An end-to-end retrieve → pack → generate pipeline grounded in a document corpus. ## Milestone 3: Q2 — Production backend depth API platform strength: keys, limits, idempotency, queues, observability. ### Concepts - [API Design](https://learn.significanthobbies.com/curriculum/concepts/api-design) — REST, gRPC, versioning, pagination. - [API Keys](https://learn.significanthobbies.com/curriculum/concepts/api-keys) — Issuing, hashing, scoping, and rotating keys for machine clients. - [Rate Limiting](https://learn.significanthobbies.com/curriculum/concepts/rate-limiting) — Token bucket, leaky bucket, sliding window. - [Idempotency](https://learn.significanthobbies.com/curriculum/concepts/idempotency) — Idempotency keys and dedup windows for safe retries of mutations. - [Retries & DLQ](https://learn.significanthobbies.com/curriculum/concepts/retries-dlq) — Backoff + jitter, poison messages, dead-letter queues. - [Observability](https://learn.significanthobbies.com/curriculum/concepts/monitoring-analytics) — Metrics, logs, traces, context propagation, semantic conventions, sampling, collectors, and telemetry pipelines. ### Build evidence - **Reusable API key system** — Issue, hash, scope, and rotate API keys for machine clients. - **Reusable rate limiter** — A token-bucket rate limiter usable as middleware. - **Job queue with retries and DLQ** — A background job queue with backoff, jitter, and a dead-letter queue. - **Structured logging package** — A structured logger with a request correlation id. ## Milestone 4: Q3 — Storage systems Build the storage foundations behind a Turbopuffer-class system. ### Concepts - [B-Tree](https://learn.significanthobbies.com/curriculum/concepts/b-tree) — The balanced, disk-friendly tree behind most relational indexes. - [LSM Tree](https://learn.significanthobbies.com/curriculum/concepts/lsm-tree) — Log-structured merge tree: write-optimized storage via sorted runs. - [Write-Ahead Log](https://learn.significanthobbies.com/curriculum/concepts/wal) — Append-only durability log written before the data pages. - [Compaction](https://learn.significanthobbies.com/curriculum/concepts/compaction) — Merging sorted runs to reclaim space and bound read amplification. - [Object Storage](https://learn.significanthobbies.com/curriculum/concepts/object-storage) — S3-style blob storage: cheap, durable, high-latency, immutable objects. - [Columnar Storage](https://learn.significanthobbies.com/curriculum/concepts/columnar-storage) — Column-oriented layout for analytics: compression and vectorized scans. ### Build evidence - **Toy write-ahead log** — An append-only WAL with crash recovery. - **Toy LSM tree** — A minimal LSM storage engine: memtable, SSTables, compaction. - **Object-storage-backed index** — Store index segments in object storage with a hot in-memory cache. ## Milestone 5: Q4 — System design and synthesis Turn the depth into architecture-level design and a real prototype. ### Concepts - [Search Platform Design](https://learn.significanthobbies.com/curriculum/concepts/search-platform-design) — End-to-end design of a search platform: ingestion, indexing, query, ranking. - [RAG System Design](https://learn.significanthobbies.com/curriculum/concepts/rag-system-design) — Architecting a production RAG system: ingestion, retrieval, generation, evals. - [Model Routing](https://learn.significanthobbies.com/curriculum/concepts/model-routing) — Sending each request to the cheapest model that can handle it. - [Agent Loops](https://learn.significanthobbies.com/curriculum/concepts/agent-loops) — The plan → act → observe loop, with memory and stopping conditions. ### Build evidence - **Search platform design doc** — An architecture doc for a search platform: ingestion, indexing, query, ranking. - **RAG system design doc** — An architecture doc for a production RAG system with an eval gate. - **Model router** — Route each request to the cheapest model that passes evals for that request class.