Glossary

The terms behind the work we ship.

Plain-English definitions of the AI concepts we use most often in client engagements. Designed to be cited — by humans, search engines, and answer engines.

RAG (Retrieval-Augmented Generation)

A pattern where a large language model answers a question by first retrieving relevant documents from your own knowledge base — and then generating an answer grounded in those documents. Solves the two biggest LLM problems for enterprises: hallucinations and staleness. Used in production for AI assistants over CRM, contracts, research libraries, and internal wikis. NxtConnect AI uses RAG (specifically GraphRAG, see below) as the foundation of its relationship-intelligence platform.

GraphRAG

An evolution of RAG that adds a knowledge graph layer between retrieval and generation. Instead of treating each document as an isolated chunk, GraphRAG links entities (people, companies, projects, deals) into a typed graph and walks relationships at query time. The result: better multi-hop reasoning ("who at company X knows person Y?") and far more accurate answers over relationship-heavy domains like sales, recruiting, and consulting. NxtConnect AI ships GraphRAG natively.

Vector Search (Semantic Search)

Search by meaning, not by keywords. Each piece of content (a sentence, a paragraph, a contact's bio) is converted into a high-dimensional vector via an embedding model; queries are converted the same way and matched by cosine similarity. Lets users ask "find contacts who work on fraud detection in fintech" and get accurate results even if those exact words appear nowhere in the source data. Foundation of most modern AI search products.

Hybrid Search

Combining vector (semantic) search with traditional keyword (BM25) search and re-ranking the merged results. Pure vector search misses exact-match queries (product codes, model numbers); pure keyword search misses paraphrases. Hybrid wins on both. Production systems at NxtConnect AI use hybrid search across structured CRM data + unstructured notes + email + meeting transcripts in a single index.

Embedding Model

A neural network that converts text (or images, audio, code) into a fixed-length vector capturing its meaning. Similar inputs land near each other in vector space. The choice of embedding model dictates the quality of every downstream vector search and RAG application — domain-specific embeddings (legal, medical, sales) outperform general-purpose ones for specialized corpora. Top open models today: nomic-embed, bge-large, jina-v3. Top closed: OpenAI text-embedding-3-large, Cohere embed-v3.

AI Agents

LLM-driven systems that can plan, take actions through tools (APIs, databases, browsers), observe results, and iterate toward a goal — without step-by-step human supervision. Different from a chatbot: an agent decides what to do next based on its current state. Production examples: outbound BDR agents that research a prospect, draft personalized outreach, and queue it for human approval; ops agents that triage support tickets across systems.

Customer-facing AI

AI applications that interact directly with your customers or prospects — chat assistants on your product, AI-powered sales outreach, support automation, onboarding agents, AI-recommended next-best-actions. Distinct from internal AI (analytics, ops automation) because the stakes are different: latency, tone, brand consistency, and failure modes are visible. Most NxtConnect AI engagements ship customer-facing AI first, because that's where the leverage is.

AI Adoption

The post-launch work of making sure an AI system actually gets used by the people it was built for. Includes change management, enablement training, measurement, feedback loops, and re-shaping the system based on real usage data. Most AI implementations fail not at the technical layer but at adoption — beautiful AI sitting unused because no one trusts it, knows about it, or has time to learn it. Adoption is what separates pilots from production.

AI Implementation

The work of taking an AI capability from idea or pilot to a production system that ships, integrates with your existing stack, and survives real load. Not the same as buying an AI tool: implementation includes data plumbing, security, evaluation harnesses, fallbacks, and monitoring. A successful implementation is invisible from the user's side and well-instrumented from the engineering side.

Salesforce Simplification

Replacing or augmenting bloated Salesforce instances with lightweight AI-native CRM that surfaces the right contact, conversation, and next action without admin overhead. NxtConnect AI's position: most teams use 8% of Salesforce, pay for 100%, and lose hours per week to data entry. AI-native CRM inverts that — auto-captures interactions from email and calendar, builds the graph automatically, and surfaces the few things that actually need human attention.

Want to talk specifics?

Book a 30-minute discovery call and we'll map these to your stack.

Book a discovery call Read insights