Let’s talk

Services / Cloud & AI integrations

Documents, queues and applied AI. Inside the workflow, not bolted on.

I connect applications to cloud services when they solve a specific problem: storing and transforming documents, moving heavy work out of the interface, reading data from documents with AI and having the people who use it confirm it.

Tell me about your projectSee the projects

When it helps

The problems I tackle.

  1. 01

    Your staff copy data from documents by hand: reports, IDs, vehicle registrations, certificates. It’s slow, and mistakes are expensive.

  2. 02

    Large files slow the application down: uploads that stall, previews that never show up.

  3. 03

    Long operations such as sending to external systems, generating PDFs and OCR keep users waiting.

  4. 04

    You want to use AI, but without made-up results and without your data leaving the European Union.

  5. 05

    Errors in production stay invisible until a client reports them.

  6. 06

    Several applications each handle files their own way, with different security rules.

What I can do

The work, in practice.

  • AI document reading

    Structured extraction with models on AWS Bedrock: every value comes with its page, source text and confidence level. A person reviews and confirms.

  • Document management

    Direct uploads to S3 with signed URLs, paths decided by the server, previews and metadata generated by a dedicated worker.

  • Asynchronous processing

    SQS queues and separate workers for OCR, PDFs, sending to external systems and large files. The interface stays fast.

  • System integrations

    Signed calls between services, idempotent webhooks, events written in the same transaction as the data.

  • Observability

    Self-hosted error monitoring, with grouping and notifications by severity.

Who I work with

I work with product teams and companies that already have an application, or are building one, and want to integrate documents, AI or cloud services reliably. I can join an existing team or take the integration from start to finish.

Training Artificial Intelligence Week by Profession AI (2024): Python, machine learning, deep learning and neural networks.

  • AWS S3
  • SQS
  • Bedrock
  • Textract
  • Node.js
  • Laravel
  • Docker
  • GlitchTip

How I work

From the problem to the release, and beyond.

  1. 01

    Understand the flow

    Where the data comes from, who uses it, what happens if it’s missing or wrong. AI goes in only where it brings a measurable benefit.

  2. 02

    Design the boundaries

    What stays in the application, what goes into a queue, who can write where. Data stays in EU regions when needed.

  3. 03

    Build with guarantees

    Structured outputs, limited retries, idempotency, errors reported. Every step can be checked.

  4. 04

    Measure and improve

    Timings, errors and uncertain fields become data for improving the flow over time.

A real example

In UnigestPro a case file can start from documents. The operator uploads the files; the worker confirms the upload and generates previews; then a model on AWS Bedrock reads every page in a single call and returns the data in a fixed structure. The rule is simple: nothing gets made up here. Uncertain fields are flagged, and the operator confirms everything before the case file is created.

The same standard for files, with direct uploads to S3 and previews generated by the shared worker, is now also used by Vidimo and Recapito.

Projects

Where I’ve done it before.

Contact

Got a product
to build?

Tell me about the problem, the context and where you want to go. I’ll reply with precise questions and a concrete proposal.

Tell me about your projectConnect on LinkedInGitHub

ciao@giovannibevacqua.com