What We Believe
Structure first, then intelligence
Value in AI starts with clean, connected data. We build the pipeline that turns messy sources into governed, AI-ready foundations.
Grounded, not guessed
Every answer carries citations and lineage back to the source. Trust comes from being able to verify — not from a model's confidence.
Data sovereignty by design
Your data stays in your environment. We index the intelligence — embeddings and graphs — not your sensitive files.
Our Vision
To make the world's unstructured data searchable, connected, and actionable — for the institutions that run society and the businesses that drive it.
Our Approach
We combine knowledge graphs, embeddings, and agentic systems into a single pipeline: structure, connect, retrieve, act.
Our Impact
From legal research to government records, we help organisations unlock the decisions that were locked inside their documents.
What We Build
The full stack that turns raw data into answers your organisation can trust and act on
Knowledge Graphs
We model your world as a connected graph — people, organisations, documents and the relationships between them — so answers join the dots across silos, with lineage you can trace.
- Entity resolution across disconnected sources
- Relationship & citation extraction at scale
- Multi-hop reasoning with full provenance
Embeddings & Vector Search
We embed your text and index it with HNSW, fusing semantic, keyword and graph search with reranking — so the right document surfaces in milliseconds.
- HNSW vector search over millions of documents
- Hybrid retrieval: semantic + keyword + graph
- Domain-tuned reranking for precision
Agentic AI Systems
We build assistants that do more than chat — they plan, decompose queries, call tools and act across your systems, with human oversight and audit trails built in.
- Multi-step tool use and query decomposition
- Human-in-the-loop with full audit trails
- Model-agnostic and cost-optimised
AI-Ready Data Foundations
AI can't reason over mess. We ingest, extract, deduplicate and enrich unstructured sources into clean, governed, AI-ready data — securely, in your environment.
- Document intelligence and OCR across formats
- Domain-aware chunking and enrichment
- Privacy-first, zero-custody architecture
Selected Work
How we put the full stack together — knowledge graphs, embeddings and agents in production
Turning 50,000+ judgments and 199,000 provisions into searchable intelligence
For legal practitioners, we structured South African case law and legislation — judgments, provisions, practitioners and judges — into a connected knowledge graph with hybrid vector search and an agentic research assistant. Practitioners find the right authority in seconds and trace every answer back to its source.