Useful AI assistants and agents, integrated into your real tools
I design and integrate AI assistants, agents and smart features into your existing applications, connected to your tools and your workflows, with a pragmatic and measurable approach.
Trust signals
- AI use cases scoped from your real processes, not from the hype.
- Integration into your existing tools (SaaS, CRM, website, back office) without rebuilding everything.
- Secure connection to your existing tools (CRM, email, calendars, internal APIs) under defined access rights.
- Tracking of usage costs (API, tokens) and of answer quality over time.
Performance
Assistants and agents: customer support, internal agents, email and ticket automation.
Product clarity
In-product AI features: content generation, classification, data extraction, document processing.
Strong foundation
Every AI project starts by identifying one to three high-impact, measurable use cases, with a clear split between essential, useful and gimmick.
Common problems
Useful AI is more than plugging in a chatbot
Approach
I start from use cases, not from the technology
Every AI project starts by identifying one to three high-impact, measurable use cases, with a clear split between essential, useful and gimmick.
The goal is a reliable integration: a scoped answering perimeter, guardrails, logging and continuous quality evaluation.
You keep visibility on usage costs, on the limits of the system and on the trade-offs, from POC to production.
Technical expertise
Common scopes on an AI project
- Assistants and agents: customer support, internal agents, email and ticket automation.
- In-product AI features: content generation, classification, data extraction, document processing.
- Integration: OpenAI / Anthropic / open-source model APIs, function calling, webhooks, MCP.
- Production: answer evaluation, cost monitoring, data security, GDPR.
Why work with me
A product approach applied to AI
- Web product and AI skills combined: AI fits into a solid application, not the other way around.
- A measurable approach: every use case has a success criterion before a single line of code.
- Transparency on what AI can do, what it cannot do, and what it really costs.
Technologies
Technologies suited to AI projects
Process
A process that proves the value before scaling it
AI scoping
Identifying use cases, available data and success criteria.
Targeted POC
A fast prototype on the priority use case to validate feasibility and quality.
Integration
Wiring into your real tools: UI, data, access rights, guardrails.
Go live
Evaluation, cost and quality monitoring, iterations based on real usage.
Frequently asked questions
Questions about AI integration
A few useful answers to scope an AI assistant or agent project.
What kind of AI projects do you deliver?
Assistants and chatbots connected to your tools, automation agents (email, tickets, documents), AI features built into an existing application (generation, classification, extraction) and upfront scoping or advisory work.
Do we need an existing application to add AI?
No. I can add AI to an existing tool or build the application and the AI layer together.
Can the assistant connect to our existing tools?
Yes: CRM, email, calendars, databases or internal APIs can be wired in through function calling, with access rights defined for each action.
How do you avoid made-up answers?
A scoped answering perimeter, guardrails on the allowed actions, test sets and continuous quality evaluation before and after go-live.
What about confidentiality and GDPR?
Hosting and providers are chosen to fit your constraints (European options or self-hosted models), with minimisation of the data sent and access logging.
How much does an AI project cost?
A targeted POC usually runs on a short format, a few days. The total cost depends on the use case and the integrations; API usage costs are estimated from the scoping stage.
Which models do you use?
Whichever fits the need: OpenAI, Anthropic or open-source models, depending on the quality required, the data constraints and the budget.
Can we start small?
Yes, and it is the recommended path: one use case, one POC, clear success criteria, then extension once the value is proven.
What happens after go-live?
Monitoring of answer quality and costs, adjustments to prompts and data, and evolutions based on real usage. A maintenance plan can cover this follow-up.
Can you train the team to use AI?
Yes, a handover is included: how the system works, where its limits are and good prompting practices.
Can you install a self-hosted assistant such as OpenClaw or Hermes Agent?
Yes. I install and secure this kind of agent for business use: VPS hosting, permissions and guardrails, WhatsApp, Telegram or Slack connection, model selection and maintenance over time.
Related services
Take the next step
Want to bring AI in without ending up with an expensive gimmick?
I can help you identify the right use cases, validate feasibility on a short POC and integrate AI into your real tools.
