Ember theme
Technology Model-agnostic Self-hosting supported

Our AI technology stack.

We do not have a house stack we sell you. We have a set of trade-offs we understand well enough to choose between — and the evaluation practice to know when we chose wrong.

Maker choosing interoperable components for an AI technology stack
HOW WE CHOOSE

Model selection is an engineering decision with four inputs, in roughly this order of weight: where the data is allowed to live, what the task actually demands, what latency the interface can absorb, and what a request costs at your projected volume. Capability is rarely the binding constraint — most production tasks are comfortably within reach of several models, and the choice is decided by the other three.

So we benchmark against your workload rather than against a leaderboard. A model that leads on public reasoning benchmarks can be beaten on your extraction task by something a quarter of the price, and the only way to know is to measure. That measurement is cheap; the assumption is not.

We also design for substitution. Model providers deprecate versions, change pricing and occasionally change behaviour within a version. A system where the model is a swappable component survives that; one where prompts are entangled with a specific model's quirks does not.

THE STACK
Frontier models
Open weights
Serving & inference
  • vLLM
  • Ollama
  • AWS Bedrock
  • Azure OpenAI
Retrieval
  • pgvector
  • Qdrant
  • Pinecone
  • Elasticsearch
Orchestration
  • Model Context Protocol
  • LangGraph
  • Temporal
  • Celery
Evaluation & observability
  • Custom eval harnesses
  • OpenTelemetry
  • Langfuse
  • Grafana

Chosen per engagement against your constraints. We are not a reseller for any of these and have no incentive to prefer one.

EVERY TECHNOLOGY PAGE
TEC.01 Model Context Protocol (MCP) What MCP is, its security model, and when a bespoke tool layer is still the better answer. → TEC.02 Multi-Agent Systems Supervisor and handoff patterns — and why most multi-agent problems are one-agent problems. → TEC.03 Evaluating AI Systems Building evaluation sets, LLM-as-judge failure modes, and gating deploys on regression. → TEC.04 Context Engineering Retrieval, compaction and memory — why prompt engineering is the wrong frame in 2026. → TEC.05 Choosing a Vector Database pgvector, Pinecone, Milvus and Qdrant compared on the axes that actually decide it. → TEC.06 Fine-Tuning & Adaptation LoRA, full fine-tuning and distillation — what each one buys, and why retrieval usually comes first. → TEC.07 Inference Optimisation Quantisation, batching, KV caching and speculative decoding — the levers that decide what serving costs. → TEC.08 AI Observability Tracing, cost attribution and drift detection for systems whose failures are silent by default. → TEC.09 Multimodal AI Vision, document and audio models in production, and where the accuracy claims stop holding. → TEC.10 Model Families Every model family we build on, what each is genuinely good at, and how the selection is made per workload. 10 PAGES TEC.11 Jev: AI Tools & Fast Decisions Build decision-making tools with TypeSafe AI Jev: parallel typed outputs, model routing, confidence gates and integration design. Includes dated provider specifications and links to published benchmarks. 4 PAGES TEC.12 Managed Agent Runtimes OpenAI Agents API public beta: persistent sessions, sandbox choices and practical enterprise pilots for research, document review and incident investigation. → TEC.13 Agent-to-Agent Interoperability A2A and MCP explained through enterprise handoffs, September 2026 platform announcements, shared context, permissions and a measured integration pilot. → TEC.14 Quantum Computing What quantum computers can and cannot do today, how you reach one through the cloud, and where the honest overlap with AI actually is. 5 PAGES TEC.15 Voice AI & Speech Systems Speech in, speech out: latency budgets, interruption handling, telephony integration and where voice agents actually break. → TEC.16 Structured Outputs & Tool Calling Function calling, JSON schemas and constrained decoding — the mechanism most production AI systems are actually built on. → TEC.17 AI Guardrails Input and output filtering, policy enforcement and escalation design, and why a guardrail is a control rather than a product. →
Keep reading
Services
AI Architectures
→
Reference
Reference Architectures
→
Compare
Comparisons
→
Trust
Trust & Security
→
30-minute technical call · no deck

Which model for
your workload?

Bring the task and the constraints. Benchmarking against your actual data beats anyone's opinion, including ours.

BOOK A TECHNICAL CALL REFERENCE ARCHITECTURES
CASE STUDIES

Shipped work.
Go and check it.

The work we can name, with the live site, our scope and the boundary made explicit. Select a project to see the evidence; each is a full case study, not a logo or a claim.

sonora.com
The Sonora homepage on desktop: a full-bleed dune landscape behind the headline “Transform Your Life with Sound”, with App Store and Google Play download buttons.
sonora.com — homepage, 1440×900 sonora.com →
Live Consumer wellness · Mobile + web

Sonora

Cognitive AI Ltd · 2026

A free sound-wellness app, described by its publisher as AI sound therapy that reads a short vocal sample at the start of a session and generates a soundscape for that moment. We designed and built the website and its backend, produced assets for the iOS and Android apps, and supported the application prototype.

Read the case study →

See every published project →

WHO WE HAVE BUILT FOR

Twenty-one years of applications, platforms and campaigns for names you know.

See all of our work →