05 · Add AIIngenara, AI feature integration

Add AI to the application you already run

Most "AI projects" stall because they start with the model and not with the application. Our AI integration work adds features to existing products that are already stable — a question-answering assistant over your documents and data, semantic search, an LLM step inside a workflow, or an MCP server that lets Claude and other assistants work with your product. Fixed scope, fixed price, and an evaluation set before any model is chosen.

Book a 30-minute triage call See our open-source MCP server

Price: Fixed, agreed before work starts
Duration: Two to four weeks
Prerequisite: A stable application

Order of work: a written success measure, then a set of real test questions, then a model chosen for your constraints, then shipped only if the agreed threshold is met.

01Four kinds

Four kinds of AI feature

Knowledge assistant (RAG)

Answers from your documents and data

Retrieval over your documents, tickets or database, with citations; an evaluation set of real questions and expected answers; guardrails; cost per query measured.

Semantic search

Search that understands what people mean

Embeddings-based search inside your product, with a keyword fallback.

LLM workflow step

An AI step inside a process you already run

Classification, extraction, drafting or routing inside an existing process, with human review where it matters.

MCP server development

A connector that lets assistants use your product

Expose your product's data and actions to Claude, ChatGPT and other assistants through the Model Context Protocol; published to the official registry if you want it public.

02Our rules

What we insist on

A written success measure before we start
An evaluation set built from real usage
Guardrails: prompt-injection defences, data-access boundaries per user, logging
A token budget with alerts
Vendor choice made for your constraints (hosted API or open-weight model), not by habit

03Not offered

What we do not do

"Add AI" with no defined outcome
Train or fine-tune models at scale
Promise accuracy numbers before the evaluation set exists
Build chatbots that replace support teams on day one

04Questions

Frequently asked questions

Not answered here? Ask on the triage call.

What does an MCP server cost to develop?

It depends on how much of your product the server exposes; read-only servers are the smallest. The scope call gives you a fixed price before anything starts.

Which LLM provider do you use?

Whichever fits your data-residency, cost and quality constraints; we build so the provider can be swapped.

Will our data be used to train models?

Not under the provider settings we use (zero-retention or enterprise terms where available); the data path is documented in the deliverable.

Can you add AI to a .NET application?

Yes — Semantic Kernel or direct API integration; the same patterns apply to Node and Next.js.

What if the results are not good enough?

The evaluation set is run before cut-over; if the agreed threshold is not met, the feature is not shipped and the reasons are in the report.