Questions clients ask before they work with us

Send a few lines through the contact page. A practice lead replies within one business day to set up a thirty-minute scoping call. If the problem is clear we propose a fixed-scope build; if it is not, we propose a short discovery sprint. Both are described under engagement models.

A discovery sprint of one to three weeks. It produces a written specification, an architecture, a risk list and an estimate you can take to anyone, including another supplier.

Yes, and we prefer it. A dedicated team joins your backlog, your repositories and your rituals, with a named technical lead accountable to you. Knowledge transfer is part of the plan from the first sprint, not an afterthought at the end.

Teknowledge Software Limited is headquartered in the UK, with delivery teams in Australia, India, Sweden and South Korea. Between the five offices we cover more than eighteen time zones, so there is almost always someone working while you are.

A discovery sprint can usually begin within two weeks of a signed statement of work. Builds and dedicated teams depend on the people required; we tell you the real date during scoping rather than promise one we cannot keep.

Yes, before any detailed discussion if you prefer. We are happy to use your template or ours. Confidentiality is also written into every statement of work.

Discovery sprints are a fixed fee. Builds are fixed price per milestone where the scope is clear. Dedicated teams are billed as monthly capacity, fractional AI CTO work as a monthly retainer, and managed services as a monthly fee with an improvement allowance. We do not publish rate cards because the right team shape changes the number; after a scoping call you receive a range with the assumptions that drive it.

Builds are invoiced against milestones set out in the statement of work, usually with a deposit at signature. Teams, retainers and managed services are invoiced monthly. Exact terms are written into each statement of work before any work begins.

You own the deliverables and the code we write for you, on payment. We keep our pre-existing tools, libraries and know-how, and we tell you in the proposal if a solution includes licensed components such as our Eventify platform.

Fixed-scope work ends at acceptance of the final milestone. Teams, retainers and managed services carry a notice period stated in the statement of work. In every case you receive the code, documentation and access credentials for what has been built.

Anthropic Claude, OpenAI, Google Gemini and open-weight models such as Llama and Mistral. Every model sits behind one interface in our applications with routing, evaluation and guardrails in our code, so the choice can change after an evaluation run. See technologies and platforms.

Only when prompting and retrieval have been measured and found wanting. Fine-tuning adds cost, maintenance and a dependency on training data; most business problems are solved sooner with better grounding and evaluation. Our guide to prompt engineering versus fine-tuning explains how we decide.

By grounding answers in your data with retrieval, requiring citations, calling live tools for facts that change, and gating low-confidence answers so the system says it does not know. Every feature ships with an evaluation set that catches regressions. The support bot case study shows the pattern in production.

Twice. Offline, against an evaluation set of real questions with expected answers, before anything ships. Online, against the business measure you agreed in discovery, such as handled conversations or hours saved. Our guide to measuring AI return on investment covers how to pick that measure.

Then discovery says so, and the first deliverable becomes a data plan rather than a model. Most AI projects that fail do so on data access, quality and permissions, not on the model. See data readiness for AI.

For a bounded use case with accessible data, typically six to twelve weeks from the start of discovery to a pilot used by real people. Enterprise integrations and compliance reviews extend that, and we say so in the estimate.

Yes. Agents that call tools, with permissions, approval steps and logging built in from the start. We build tool layers as MCP servers where it helps, so the same tools serve your assistant, your automations and your own staff. See AI agents and automation.

Wherever you need it to be. We build inside your cloud accounts where possible, choose regions to meet your residency requirements, and run open-weight models inside your network when data cannot leave it. Model providers are chosen with the same constraints.

No. We do not train models on client data unless a statement of work explicitly says so, and we configure commercial model APIs with training and retention options switched off where the provider offers them.

Yes. Open-weight models, vector databases such as pgvector and Milvus, and our application code all run on your infrastructure. Our guide to on-prem versus cloud LLMs sets out the trade-offs.

Yes. We classify your systems under the Act, prepare the technical documentation and set up the controls an ISO 42001 management system expects. See EU AI Act readiness and ISO 42001.

Yes. An LLM security audit covers prompt injection, data leakage, access control and supply chain, with a prioritised fix list. Red-team exercises are available for higher-risk systems.

Named accounts with the least privilege needed, multi-factor authentication, access logged, and everything revoked at the end of the engagement. We never share credentials between people and we document every access we hold.

Yes, greenfield implementations of Sales, Service, Marketing and Experience Cloud, and re-implementations of orgs that have become hard to change. See Salesforce implementation.

Yes. We implement Agentforce and Einstein features where they pay off and connect external models where they do not. A Salesforce AI roadmap is the usual starting point.

Yes, through managed services: administration, release management, user support and a monthly improvement allowance, with agreed response targets.

Yes. We profile the data first, clean it with you, migrate in rehearsals and reconcile every record before cut-over. See data migration.

If your events need registration, agenda, networking, check-in and analytics under your brand, Eventify as a white-label platform is faster and cheaper. If your requirements are unusual enough that configuration cannot express them, we build custom. Discovery tells you which.

Yes. Branded apps in the app stores, a web experience on your own domain, your colours and your content. Organisers configure events themselves in the admin console.

Yes, where agreed in the statement of work, typically for check-in, badge printing and kiosk setups at larger events. Remote support covers everything else, across time zones.

Yes. Attendee data can be imported, synchronised through the API or pushed by webhook from most registration tools. We confirm the integration path during scoping.

A shared channel for day-to-day questions, a short weekly call, a demo of working software every two weeks, and written minutes after every meeting. You can talk to any engineer on the team directly.

Yours, wherever possible: your repositories, issue tracker, chat and cloud accounts. Where you have no preference we bring our own and hand them over at the end.

A decision-maker who can attend the fortnightly demo, access to the systems and data in scope within the first week, and one place for questions. Most delays we have seen were waiting, not building.

A warranty period is included in every build. After that, managed services or a dedicated team keeps the system monitored, evaluated and improving. Our delivery process describes the full lifecycle.