Plotting the First Point: What an Embedding Actually Is
why a surveyor's first job is turning a real place into a coordinate, and how an embedding does the same thing for meaning.
Part 1
Part 2
Why Nearby Plots Share a View: Similarity and Distance Metrics
why two plots close together on a map tend to share a real view, and how distance metrics turn that intuition into a precise measurement between embeddings.
Part 3
Choosing the Right Instrument: Dimensionality and Embedding Model Selection
why a survey of a city block and a survey of a mountain range need different instruments, and why choosing an embedding model means choosing a dimensionality to match the job.
Part 4
The Benchmark Marker: Normalization and Reference Points
why every survey needs a fixed benchmark marker to stay honest over time, and why embeddings need normalization to stay comparable.
Part 5
Walking to the Nearest Marker: Nearest-Neighbor Search
why finding the closest surveyed marker to a given point is the whole point of having a map, and how nearest-neighbor search does the same for embeddings.
Part 6
Searching the Whole County by Sundown: Approximate Nearest-Neighbor Search
why a survey team can't walk every acre before sundown, and how approximate nearest-neighbor algorithms trade a little accuracy for a lot of speed.
Part 7
Layering the Survey: Hybrid Dense and Sparse Search
why a good survey layers a terrain map on top of a property registry, and why the best vector search systems combine dense embeddings with sparse keyword signals.
Part 8
When the Map Goes Stale: Embedding Drift and Re-Embedding
why an old survey map slowly stops matching the real terrain, and why embeddings need re-generating as models and content evolve.
Part 9
Different Maps for Different Trades: Domain-Specific and Fine-Tuned Embeddings
why a utility surveyor and a wildlife surveyor draw genuinely different maps of the same land, and why fine-tuned embeddings serve specific domains better than general-purpose ones.
Part 10
The Index That Saves the Legwork: Vector Indexes in Practice
why a well-organized survey office keeps a card index instead of re-searching every filing cabinet, and how vector indexes do the same for embeddings at production scale.
Part 11
Folding the Map Down: Quantization and Dimensionality Reduction
why a surveyor carries a folded, simplified field map instead of the full master copy, and how quantization shrinks embeddings without losing what actually matters.
Part 12
Only the Plots Zoned Right: Metadata Filtering Alongside Vector Search
why a buyer wants the nearest lakeside plot that's also zoned residential, and how metadata filtering combines with vector search to answer both requirements at once.
Part 13
Keeping the Survey Current: Upserts, Deletes, and Versioning
why a survey office needs a real process for updating and retiring records, not just adding new ones, and why vector databases need the same discipline.
Part 14
Surveying the Whole County: Clustering and Embedding Space Analysis
why stepping back from individual plots to study the whole county's patterns reveals things no single-plot survey ever could, and how clustering does the same for embeddings.
Part 15
The Plot Listed in the Wrong Place: Outlier and Anomaly Detection
why a plot recorded far from where it should genuinely sit is worth a second look, and how outlier detection in embedding space catches the same kind of mistake.
Part 16
Not Every Plot Needs a Theodolite: When Vector Search Is Overkill
why a small, simple lot doesn't need a full precision survey, and why not every search problem genuinely needs a vector database.
Part 17
Digitizing the Old Paper Deeds: Migrating Existing Data Into Embeddings
why a county's old paper deeds have to be genuinely re-surveyed, not just scanned, to join a modern coordinate system, and what migrating existing text into embeddings actually requires.
Part 18
The Drone That Surveys While You Sleep: AI-Assisted Embedding Pipelines
why a fleet of autonomous survey drones can keep a whole county's map current without a human walking every acre, and how AI-assisted embedding pipelines do the same for a growing content collection.
Part 19
When the Builder Just Says 'Something Like This': RAG and AI Agent Retrieval
why a builder can describe a plot in plain language and trust a well-organized survey office to find genuine matches, and how retrieval-augmented generation lets AI agents do the same over real, grounded facts.
Part 20
One Coordinate System, Every Search
the first point plotted, the benchmark marker, the folded field map, the drone fleet, and the builder's plain-language request, every article's lesson reassembled into one coherent, trustworthy coordinate system.