ABAdrien Bonvallet

Services

Practical automations, built to fit your process and shipped to production.

Four services that combine: most engagements start with a document flow to automate, then extend to the interface that makes the data usable.

Your inbox, handled on its own

Every incoming email gets read, understood and sorted automatically. The right actions fire without anyone opening the mailbox.

  • Structured data extraction (typed JSON) from free-form text
  • Intent classification and routing to the right team or system
  • Action triggers: ticket creation, CRM update, notification, auto-reply
LLMMicrosoft Graph APIFunction callingAsync queue

PDFs in, clean data out

Invoices, quotes, contracts: documents are read and turned into usable data. No manual re-entry.

  • OCR for scanned files, native parsing for digital PDFs
  • Key fields extracted: amounts, dates, VAT, parties, line items
  • Consistency checks and confidence scoring, with human review on doubt
OCRVision LLMSchema validationAPI / CSV export

Data your clients actually read

A clean interface where your clients and teams see what matters, instead of a raw export to decode.

  • Custom dashboards shaped around your clients' real workflows
  • Filters, search and exports, with role- and org-based access
  • Live updates wired straight to your processing pipelines
Next.jsReactPostgreSQLRole-based auth

From idea to app in production

The full product, built end-to-end: back end, interface, deployment and support once it is live.

  • API design and database modelling
  • Polished web and mobile interfaces, responsive and accessible
  • Deployment, CI/CD, monitoring and handover to your own team
TypeScriptNode.jsReact NativeCI/CD

Included in every engagement

Code and data are yours

Git repository, documentation and access delivered in your name. No dependency on me to keep running the tool.

Fixed scope and price

Each step comes with a written quote and a defined deliverable. No surprise invoice at the end of the month.

Human in the loop

Uncertain extractions are flagged and routed to a human instead of silently writing bad data.

Confidentiality

NDA signed on request, GDPR-compliant handling, European hosting by default.

A four-step method

Every step has a deliverable and a price agreed up front. You can stop at the end of any of them.

  1. 015 days

    Diagnostic

    I observe the real process, measure the time spent and map the flows. Deliverable: a costed scoping note with expected gain and implementation cost.

  2. 022-3 weeks

    Prototype

    A first working version on your real data, tested by your team. Goal: validate accuracy before investing further.

  3. 032-4 weeks

    Production

    Integration with your tools, edge-case handling, consistency checks, logging and manual fallback whenever confidence is low.

  4. 04ongoing

    Support

    Monitoring, adjustments and new features. The code and documentation are yours: your team can take over at any point.

Let's talk about your process

Thirty minutes to understand your document flow and tell you, plainly, what can be automated and what cannot.