Every project we have shipped has its own story here, alongside guides and perspectives on building AI that ships.
Selected projects written up in full: the problem, how it was built, and what changed once it was running.
An agentic outbound pipeline for a Wall Street brand selling to private equity firms, hedge funds and family offices.
4 min read
An entity-linked knowledge graph over documents, with multi-hop reasoning, auditable citations and Salesforce integration, built for import and export compliance.
5 min read
A multi-agent framework that synthesizes respondent data for text and image studies, benchmarked against live panels.
5 min read
An IVR with a real-time AI video consult for pre- and post-procedure guidance, with automatic logging and scheduling.
3 min read
Diffusion-based try-on with conversational sizing and retrieval over the product catalogue, wired to live inventory, for a leading clothing brand.
3 min read
Data ingestion that keeps going when a source breaks, gathering product data, reviews and listings from multiple sources for a luxury fragrance business.
3 min read
A knowledge graph of routes, hubs and service levels, with optimization and exception handling exposed through APIs.
3 min read
Anomaly detection on vibration and temperature for industrial conveyors, combined with retrieval and forecasting for planners.
3 min read
A production-scale retrieval platform over millions of documents across many client deployments, with multimodal search and a mobile chatbot.
4 min read
Domain-specific model tuning with LoRA and Q-LoRA, built as a repeatable framework rather than a one-off experiment.
3 min read
Gradient-style optimization over prompts and pipelines, tuning systems against measurable objectives instead of intuition.
3 min read
Clusters raw conversation logs into coherent intents without labelled data, so support taxonomies come from real usage.
3 min read
Named entity recognition tuned for messy real-world documents, with evaluation harnesses and human review loops.
3 min read
Usage dashboards and billing built into a client platform, with metering, invoicing and role-based access.
3 min readA sales assistant that talks to shoppers, answers what they actually asked, and keeps the team out of the repetitive parts.
What is actually changing in AI at work, from systems that act on their own to the rules companies are writing for themselves.
What language models are really doing to customer service, and where they still need a person.
What it takes to build AI that is fair, that you can explain, and that somebody is answerable for.
How to get useful data out of documents that arrive in every format imaginable, and what goes wrong along the way.
What happens to a model after it leaves your laptop, and how to keep it working once it is out there.
How to tell whether an AI project was worth doing, using examples from work we have shipped.
How to use AI responsibly and stay on the right side of data privacy rules.
Tell us the problem and what needs to change. We will say honestly whether AI is the right instrument for it, or something else and how we would approach the solution.
