AI automation · Cluj-Napoca, Romania · working remotely
I design and build AI workflows that read the inbox, process the documents, answer the internal questions and move data between the tools you already use. A human stays in the loop where the risk is real. The result is measured in hours removed, not in demo quality.
01 / Where the hours go
Most teams do not need another chatbot. They need one repetitive workflow to stop consuming a person’s week.
The same copy-paste, status update and data entry happens in every role. Nobody can point to the hours because they are spread across a dozen small tasks.
Measure the hours 02 / Knowledge and supportThe policy is in a PDF, the precedent is in a ticket, and the person who knows is in a meeting. New joiners ask the same questions for weeks.
Scope the assistant 03 / Documents and inboxesInvoices, contracts, forms and attachments arrive in different formats. Someone opens each one, types the same fields into another system and chases the missing ones by email.
Audit the documents 04 / Integrations and handoffsThe CRM, the helpdesk, the ERP and the spreadsheet each hold part of the truth. The integration is a person copying between tabs, or a script nobody owns.
Map the integrationWho it is for
The common thread is not industry. It is a workflow that runs often enough to measure, with someone who can say what “good” looks like.
Requests, updates and approvals move between a form, a sheet, an inbox and a CRM, and someone stitches them together.
Your team answers questions that are already written down somewhere, but the answer takes ten minutes to find.
Invoices, contracts, forms and attachments arrive in different formats and are keyed in by hand.
The process lives in your head, and every new customer adds another manual step. You need the first automation, not a platform.
02 / Services
Six services, each with a defined deliverable and a fixed price. Start with the audit if the workflow is not yet clear.
A five-day audit of one workflow: map the steps, quantify the hours, price the exceptions and decide whether to automate.
One end-to-end automation with AI at the decision point, connected to your existing tools, with a human approval step where the risk is real.
A pipeline that reads the inbox or document folder, extracts the fields, validates them and routes the result, with a review queue for the uncertain cases.
An assistant over your own documents and tickets, with citations, permissions and an evaluation set, so it is useful before it is impressive.
Connect a model to your existing systems with authentication, validation, error handling and a clear boundary around what it can change.
Monthly monitoring, exception review and a small improvement block, so a live automation keeps saving time instead of quietly failing.
03 / Evidence
Delivery and research, with the boundary of every claim written down. The automation practice is new; the engineering behind it is not.
Owned the AI and application layer, and served as the hospital’s primary technical point of contact for the AI and application scope.
Deployed NVIDIA AIQ on NVIDIA DGX Spark and extended it with tools and agent capabilities, on customer-owned hardware.
Built during an AI internship: a platform that takes a repository and agentically deploys it to a local environment, Docker or Azure.
04 / Credentials
Most workflows stop at an API call. Some need a private model, a fine-tune or a deployment on your own hardware. That is when the NVIDIA certification and the delivery record matter.
The advanced NVIDIA credential for generative AI with large language models. It covers model optimization, fine-tuning methods, distributed training, deployment and production monitoring at a depth that goes past prompt-level work.
The foundational NVIDIA credential for generative AI with large language models. It covers the vocabulary and mechanics behind LLM applications: how models are used, prompted, grounded with retrieval and evaluated.
Certification topics were studied and examined. They are not claimed as production experience. Every hands-on claim on this site links to the project that supports it. The full exam blueprint and the hands-on evidence are on the certifications page.
05 / How it runs
No open-ended discovery. One workflow, a baseline, a fixed scope and a handover.
One workflow, who does it, how often and what breaks. If the workflow is too small to automate, I say so on the call.
Five days to map the process, measure the hours and exceptions, and produce a costed design. The audit is credited against the build.
One workflow, connected to your tools, with validation, a human review step and a visible failure path. Measured against the baseline.
Before/after numbers, logs, a runbook and a session with the team. Optional retainer for monitoring and monthly improvements.
I work in English, Romanian and Italian, from Cluj-Napoca (EU time zone), with teams across Europe and the US.
06 / Questions
Short answers to the commercial and technical questions that usually decide whether a workflow gets automated.
The best candidates are repetitive, high-volume steps with a clear input and output: document and invoice processing, inbox triage, data entry between systems, internal question answering, report generation and first-line support. The work is to find the exceptions before automating the happy path.
A workflow audit is €1,500 to €2,500. One automation build is €3,000 to €6,000, document and inbox automation is €2,500 to €5,000, an internal assistant is €4,000 to €8,000, and an integration path is €3,000 to €7,000. The retainer is €750 to €1,500 per month.
No. The workflow is built around the systems you already use, through an API, webhook or structured export. Replacing a working tool is usually a bigger project than automating the step around it.
On the same samples before and after: time per item, manual touches, cycle time and exception rate. The report shows a range with the assumptions written down. It does not claim a generic productivity percentage.
Yes. I hold the NVIDIA Certified Professional: Gen AI LLMs (February 2026 to February 2028) and the NVIDIA Certified Associate: Gen AI LLMs (November 2025 to November 2027). The engineering depth matters when the automation needs a private model, fine-tuning or a deployment on your own hardware.
Send a short description of one workflow, who does it today and how often. If it looks like a fit, we have a 30-minute call and I send a one-page proposal with scope, deliverables, price and a baseline. Nothing is billable until you approve it.
Next step
The first call is free. Bring one workflow, its owner and a rough volume. I will tell you whether an audit is worth it.
Prefer email? boroicamarius01@gmail.com