A short, plain-English glossary of the AI terms that come up when buying an implementation engagement — RAG, voice clones, AI avatars, AI agents, vector databases, and the rest. Written for buyers who want to understand what they're paying for without sitting through a vendor pitch.
A firm that designs, builds, and operates custom AI systems inside a specific business end-to-end and on retainer. Distinct from an AI consultancy (which delivers a roadmap) and an AI software vendor (which sells a license). Standin is an AI implementation agency for businesses doing $1M+ in revenue. Full definition →
A cloned AI avatar — face, voice, and knowledge — that conducts live B2B discovery presentations autonomously. Replaces the introductory and qualification calls a senior salesperson currently does manually, while keeping the script under human approval. Architecture breakdown →
A photorealistic video rendering of a person — typically generated by platforms like Tavus or HeyGen — lip-synced to whatever audio it’s given. The avatar is the cosmetic layer of an AI sales presenter; the intelligence behind what the avatar says is a separate component.
A synthetic version of a specific person’s voice, trained from 20–60 minutes of clean recorded audio, capable of producing speech indistinguishable from the original at conversational length. ElevenLabs is the current standard. Paired with an avatar so spoken output and lip-sync stay aligned.
A technique where a language model is grounded in a custom knowledge base by retrieving relevant passages at query time and feeding them into the model’s context. RAG is how AI systems answer questions about a specific business without retraining the underlying model. The retrieval layer matters more than the model itself.
Storage optimized for semantic search over text embeddings — the data structure that makes RAG retrieval fast. The vector database stores the firm’s materials (call transcripts, documents, FAQs) as numerical embeddings; queries are matched by semantic similarity rather than keyword match.
The process of further training a pre-trained language model on a smaller, domain-specific dataset to shift its behavior or style. Useful for narrow, repeatable tasks; rarely necessary for most business use cases, where RAG plus prompt engineering reaches the same outcome with less cost and operational complexity.
A system in which a language model is given tools (search, code execution, API calls) and a goal, and reasons step-by-step to complete a task without explicit human direction at each step. In production B2B contexts, fully autonomous agents are constrained by scripted guardrails — improvisation is reserved for narrow, low-risk steps.
An AI system that handles voice phone calls — inbound reception, outbound qualification, scheduling, or routing — trained on a specific business’s products, services, and objection-handling scripts. Distinct from a chatbot in that it operates over telephony and natural speech.
Architecture pattern where slide narration is delivered verbatim from a pre-approved script, not generated live. The single most important guardrail in an AI sales presenter — without it, the system will eventually hallucinate something embarrassing in front of a prospect.
The subsystem in an AI sales presenter that detects when a prospect interrupts a presentation, pauses the script, answers the question via RAG, and resumes narration at the right point without repeating prior content. Sounds trivial. It is the hardest engineering challenge in an autonomous presentation system.
A focused 1–2 week engagement ($7,500 at Standin) auditing a firm’s sales process, communication workflows, internal operations, and tech stack — producing a prioritized AI roadmap with estimated ROI and a phased build sequence. Standalone deliverable and the recommended next step for new clients before committing to a build.
A monthly retainer tier where the implementation agency acts as the client’s embedded AI department — monitoring performance, tuning systems, identifying new opportunities, and replacing all individual service retainers. Natural landing point for clients running three or more active AI systems built by the agency.
The structured collection of a firm’s materials — recorded calls, presentations, FAQs, product docs, internal training material — that an AI system uses to answer questions. Built once during deployment, kept current under the operating retainer.
Definition, scope, and the four phases of an engagement.
Read the full definition →Where most of these terms (avatar, voice clone, RAG) come together.
Read the architecture breakdown →What works, what to skip, by industry.
Read the category guide →The discovery consultation translates these concepts into a written proposal specific to your business. $1,200, applied toward the audit or setup fee at signing.