LLM integration · retrieval over internal documents · agents · evaluation
We build AI into the systems a company already runs — with the sources attached, the failure cases measured, and a description of the data flow that your security and compliance people can actually read. If a task is better solved without a model, we say so in the framing stage.
A model is the smallest part of the work. The rest is the seam into your ERP, CRM, ticketing or document store.
Typical tasks
Answers grounded in your own corpus — policies, contracts, manuals, tickets — with retrieval quality measured separately from writing quality.
Typical tasks
Multi-step work with tools: reading a queue, calling internal APIs, filling forms, escalating to a person.
Typical tasks
A model that answers fluently is not a model that answers correctly. Quality is measured on your own hard cases, release after release.
Typical tasks
Where the data lives, what leaves the perimeter, what is retained and what is logged — written so compliance can review it.
Typical tasks
Models and documents change. Either a clean hand-over with documentation and monitoring, or a monthly engagement that keeps quality measured.
Typical tasks
Lengths are the usual range for a single process with a cooperative data owner. They are not a quote: the framing stage exists to replace them with a real plan for your case.
01 · 1–2 weeks
One process, written down as it happens today: the volume, the exceptions, the people involved and the cost of a mistake. The stage ends with a decision on whether AI is the right tool at all — sometimes it is not.
02 · 1–2 weeks
Where the documents live, who may see what, what may leave the perimeter, which model providers are acceptable, what the auditors and the regulator will ask for.
03 · 3–6 weeks
A narrow, real slice in production conditions: retrieval, prompting, an evaluation set from your own cases, and an interface inside the tool people already use.
04 · 2–4 weeks
The failures found in the pilot become the regression suite. Permissions, logging, rate limits, refusal paths and hand-over to a person are tightened before anyone depends on it.
05 · 2–6 weeks
Wider release, training for the people using it, monitoring and alerting, and documentation for whoever runs it after us.
06 · ongoing
A fixed monthly engagement: monitoring, re-evaluation after model or document changes, incident handling and small changes, with a named engineer.
Where the system runs is decided by the data, not by preference: your cloud tenant, your own hardware, or ours. The same build goes to all three, so the constraint can change without a rewrite — and the evaluation suite runs against whichever one you choose.
Models
Models from the main providers for hosted work, open-weight models inside your own environment where data may not leave it. The choice follows the constraint.
Retrieval
Vector and keyword search over your documents, hybrid ranking, permission-aware filtering, citation back to the source page.
Orchestration
Explicit pipelines and tool-calling agents with step logging, retries, timeouts and a defined hand-over point to a person.
Evaluation
Labelled question sets from your own operation, retrieval and answer metrics tracked per release, failure cases kept as tests.
Integration
APIs and web services against the systems you already run; no rip-and-replace, and no second copy of your data where it can be avoided.
Delivery
Containerised services in your cloud account or ours, infrastructure as code, separate test and production environments.
Observability
Per-request logs of prompt, retrieved passages, model version and outcome, retained according to your policy rather than ours.
Access control
Your identity provider, your groups and roles; the assistant never has wider access than the person asking.
Data handling in one paragraph
Your documents and records stay in the environment you choose: your cloud tenant, your own servers, or ours if you prefer and the classification allows it. We describe what leaves the perimeter, keep the audit trail of prompts and retrievals, and do not use your data to train anything. Where a requirement forbids external model calls, we run open-weight models inside your perimeter — with the accuracy trade-off stated in writing.
Two to four weeks to turn an idea into a written scope, a data inventory and a go/no-go decision. Fixed price, no commitment afterwards.
fixed fee, quoted per scope
One process, one user group, quality measured on your own cases. Ends with a working slice and the numbers needed to decide about production.
fixed fee per stage
Hardening, permissions, monitoring, rollout and documentation, delivered in stages with a demo at the end of each one.
fixed fee per stage
A monthly engagement after launch: monitoring, re-evaluation, incident handling and small changes.
monthly retainer
Every stage is scoped and priced in writing before it starts, and every stage ends with something you can read. Stopping after any stage is a legitimate outcome.
Tell us what has to change, which systems are involved and who would use the result. If an NDA is needed before details are shared, say so in the form and we will sign yours or send ours first.
Discuss a project
Tell us what has to change in the process, what systems are involved and who would use the result. We reply within one business day; an NDA can be signed before any detail is shared.
Prefer email? [email protected] · or call +1 (555) 010-0600.