The full GenAI capability library

Secure, grounded generative AI on your own data — from prompted assistants to retrieval-grounded answers to fine-tuned, private models.

techniques

disciplines

capabilities

the whole spectrum

prompted → adapted / trained · not just the fashionable end

32 prompted51 grounded (RAG)59 adapted / trained

/ browse

The 5 capabilities, in depth

Pick a capability — its plain-English job, the questions it answers, and the full toolbox scroll alongside.

the colour is the tierpromptedgrounded (RAG)adapted / trained

Retrieve & ground

answers from your data, with citations

39techniques · 3 disciplines

9 prompted16 grounded (RAG)14 adapted / trained

Point the model at your documents, policies and records so every answer is drawn from your own knowledge — and every claim can be traced back to the source it came from.

answers questions like

→“Can our people get a trusted, sourced answer instead of hunting through a hundred documents?”
→“How do we make the AI cite where each fact came from — so we can defend it?”
→“Can it search across everything we know, and still find the one right passage?”

the toolbox — 39 techniques, prompted to adapted / trained

Retrieval & chunking

15 techniques

Keyword / boolean searchTF-IDFBM25Fixed-size chunkingSliding-window chunkingSemantic / recursive chunkingSentence-window retrievalParent-document retrievalMetadata filteringContextual retrieval (chunk-prefixing)Multi-hop retrievalGraphRAGHyDE (hypothetical document embeddings)Query rewriting / decompositionSmall-to-big / auto-merging retrieval

Vector & hybrid search

14 techniques

Brute-force / flat searchCosine similarityEmbeddings (OpenAI / Cohere / BGE / E5)HNSWIVF-PQFAISSpgvectorMilvusQdrantHybrid BM25 + denseReciprocal Rank Fusion (RRF)SPLADE (learned sparse)ColBERT (late interaction)Matryoshka / binary-quantised embeddings

Ranking & grounding

10 techniques

Top-k selectionSimilarity thresholdCross-encoder re-rankCohere RerankMaximal marginal relevance (MMR)Citation / source attributionContextual compressionGrounded generation (attributed answers)recall@k / MRR / nDCG evaluationRetrieval fusion across sources

/ the point

Grounded, or it doesn't ship

Ungrounded GenAI hallucinates — so we ground every answer in your data with citations and guardrails, and prove it holds up, before it ever faces a customer.

Back to the service

agent orchestration lives with Agentic AI · train-on-your-data models with Data Science & ML · serving & monitoring with AIOps / MLOps