Answers

AI, defined plainly

The vocabulary you'll meet when evaluating an AI project — each term explained in a couple of sentences, without the vendor gloss.

Optical character recognition(OCR)

Optical character recognition is the conversion of text in an image — a scan, a photograph, a fax — into machine-readable characters. Modern OCR combines computer vision with language models, which lets it resolve ambiguous characters from surrounding context rather than reading each glyph in isolation.

Intelligent document processing(IDP)

Intelligent document processing is the automated handling of documents end to end: reading them with OCR, classifying them by type, extracting the fields that matter, validating those fields against business rules, and routing the result. It differs from plain OCR by understanding what a document is, not only what it says.

Retrieval-augmented generation(RAG)

Retrieval-augmented generation is a technique where a language model retrieves relevant documents from an external source at query time and uses them to compose its answer. Because the knowledge stays outside the model, it can be updated, permissioned, and cited without retraining anything.

Fine-tuning

Fine-tuning is the process of further training an existing model on a specific dataset so it adapts to a particular style, format, or domain. It changes the model's weights, which makes it well suited to teaching behaviour and poorly suited to teaching facts that change.

LLM evaluation(evals)

LLM evaluation is the practice of measuring a language model system's output against a fixed set of test cases with known-good answers. Because the same input can produce different outputs on different runs, evaluation is the only reliable way to tell whether a change improved a system or quietly broke it.

AI agent

An AI agent is a system that pursues a goal across multiple steps, deciding which tools or APIs to call and adapting based on the results. The distinction from a chatbot is action rather than conversation: a chatbot answers a question about an order, while an agent looks it up, updates it, and confirms the change.

Human in the loop(HITL)

Human in the loop is a design pattern where an automated system routes uncertain or high-stakes cases to a person instead of deciding alone. The system reports a confidence score with each output, and cases below a set threshold go to a review queue rather than straight through.

Hallucination

A hallucination is output from a language model that is fluent, confident, and false. It arises because such models are trained to produce plausible continuations of text, not to verify claims — so a fabricated citation is generated by the same mechanism as a correct one, and looks identical.

Vector database

A vector database stores text, images, or audio as embeddings — numeric representations of meaning — and retrieves items by similarity rather than exact matching. It is the component that lets a search for "time off policy" return a document titled "paid leave guidelines" with no shared keywords.

Model drift

Model drift is the gradual decline in a deployed model's accuracy as real-world data diverges from the data it was built against. The model does not change; the world does — new document layouts, new phrasing, new customer behaviour — and performance degrades quietly rather than failing outright.

Straight-through processing(STP)

Straight-through processing is the share of transactions a system completes end to end with no human intervention. It is the headline metric for document and workflow automation: a pipeline at 80% STP handles four of five items automatically and routes the fifth to a person.

Got an idea? Let's make it real.

Tell us about your problem. We'll come back within one business day with a take, a rough plan, and a call invite.

Schedule a call Book a 30-min intro call Email us