AI-Ready Audit
Diagnostic — Entry product2 to 3 weeks · fixed fee, billed upon order
From €1,500
100% billed upon order
Labs works on that lock, and nothing else: maturity diagnostic, data layer, agents wired to your real sources — then the measurement of what they change. For executives who want to move from POC to production.
In 2026, virtually every SME uses generative AI on a daily basis. Yet almost none have connected it to their own data. The result is the same everywhere: generic answers, hollow agents, data siloed across Drive, CRM, mail and invoices.
The
bottleneck
isn't
the
model
choice.
The
bottleneck
is :
how
to
make
an
SME's
data
usable
by
AI,
in
a
contextual,
secure
and
maintainable
way.
LinkTec
Labs
answers
that
bottleneck.
From diagnosis to deployment, through to long-term advisory.
2 to 3 weeks · fixed fee, billed upon order
6 to 12 weeks · quoted to scope, in milestones
6-month minimum · monthly retainer
Every Labs engagement follows the same rigorous protocol — from diagnosis to continuous optimisation.
Map the data, identify quick-ROI use cases, measure AI maturity.
Build the data layer: ingestion, vector database, connectors, custom agents.
Deploy to production: agent orchestration, business integrations, continuous monitoring.
Measure, adjust, evolve: KPI steering, prompt A/B testing, continuous R&D.
Five steps, always the same. Each one is a place where data can leak, be badly filtered or become unverifiable — so a place to ask any provider a precise question, LinkTec included.
Documents, databases, e-mails, business tools. The scope is set in writing: whatever is not in it is never copied.
Extraction, chunking, then a description that places each fragment back in its source document — without it, an extract loses its meaning.
Semantic and lexical search, merged. Permission filtering applies to the search itself, not afterwards.
The model only receives the fragments retained. Every answer points back to the passages behind it, and the trace is exportable.
A reference question set measures faithfulness to the sources. Any change is justified by a figure, not by an impression.
Overview. The engineering detail — permission filtering at search time, citations, audit log, evaluation, end of engagement — is published on Security & architecture. This is a reference architecture: an engineering target and a scoping method, not an estate already deployed across clients.
LinkTec Labs combines university education in artificial intelligence & data in progress with hands-on experience in robotics and AI research laboratories.
This dual culture — academic and operational — brings to SME problems the same methods proven in research: rigorous modelling, experimental validation, documented architecture choices.
What follows describes technical skills and how they fit together, not a list of products to order. The bricks retained, the model provider and the hosting region are settled with you at scoping — nothing is imposed by default.
PDF / DOCX / spreadsheet extraction, chunking, provenance metadata. Every fragment keeps the context of its source document.
Semantic and lexical search, merged. Pinecone · Qdrant · Weaviate, or a self-hosted index when the scope calls for it.
Re-ranking, mandatory citations, permission filtering applied at search time — not after generation.
Claude API and business tools called by one agent: typed tool schemas, error recovery, token budgets, output guardrails.
n8n, webhooks, queues, idempotent and replayable runs: a job re-run twice does not produce two effects.
Reference question set, measurement of faithfulness to sources and of retrieval relevance, run logging.
Commissioning, reverse proxy, secret management, backups, monitoring and logs. What holds in production, not just in a demo.
Modelling and simulation of a business process, to test a decision before applying it for real.
Structured prompting and guided examples first; supervised adaptation of a model (fine-tuning) only when style, format or a very specific business vocabulary cannot be obtained any other way. In the vast majority of cases, putting data in context gives a better result at a lower cost — that is what is recommended by default.
A scoped engagement in its own right, inside level 2: a test set defined before starting, a measured comparison against contextualisation alone, and the move to production decided on that result. No performance gain is promised in advance, and the cost depends on the dataset — it is settled at scoping.
NDA before any exchange, DPA when personal data is processed, alignment with GDPR & the EU AI Act, hand-back in open formats and deletion.
For a private document assistant, the framework connects authorised sources, contextualised ingestion, a hybrid index, retrieval-time access control, the model selected at scoping, citations, an audit trail and evaluation. It is a reference architecture, not infrastructure claimed to run for several clients.
NDA before data exchange, DPA where required, named hosting and subprocessors, access restricted to authorised contributors according to their role, and explicit retention. Dedicated infrastructure, at-rest encryption and detailed audit logging are configurable, never assumed.
A maturity framework designed to help SME executives benchmark themselves against their sector — and know exactly where to act first.
An AI-Ready audit is a structured diagnostic of an SME's AI maturity across 6 dimensions: data, processes, tools, governance, skills and culture. It results in a 30 to 50-page report including an AI-Ready score, a data mapping, a prioritised list of 5 to 10 quick-ROI use cases and a 6/12/24-month implementation roadmap. The goal is not to sell a solution — it's to provide a clear view of the current state and action priorities.
A traditional AI consultant delivers strategic recommendations without building the system behind them. LinkTec Labs combines both: the audit leads to concrete technical deliverables (ingestion pipelines, vector databases, custom agents) that can be put into production by the Labs team itself. Diagnosis and execution are handled by the same technical capability — not a consultant handing off to an integrator.
No. The vast majority of SMEs hold unstructured data (Drive, mail, invoices, partial CRM, Word documents, manual exports). The AI-Ready audit specifically includes the mapping and quality assessment of these sources. Level 2 (data layer) then builds the ingestion and contextualisation pipeline to make this data usable by AI — without requiring a full IT overhaul.
Yes. When the engagement involves access to internal or sensitive data, a non-disclosure agreement (NDA) is signed before any document exchange. For a smoother start, it is also possible to work initially on a restricted scope or on anonymised data. In all cases, access to information is strictly limited to the operational needs of the engagement, logged, and the data is returned or deleted at its end.
An audit does not guarantee ROI — it identifies the use cases where ROI is probable and prioritises them according to implementation effort. Each retained use case is documented with its expected gain hypothesis, estimated implementation cost and success conditions. Actual ROI is measured later, once in production — not before. This transparency around uncertainty is part of the research-driven approach: nothing is promised that cannot be measured.
Three possible next steps, with no commitment: (1) the audit stands on its own — the executive uses it to steer internal decisions; (2) data layer implementation (level 2) — building the infrastructure and first agents over 6 to 12 weeks; (3) a monthly retainer (level 3) — long-term support with AI watch and targeted R&D. The choice is made after reading the report — not under the pressure of a sales signature.
Let's discuss your project in 30 minutes. No commitment.