Sergiu Wittenberger
Founder - Lead AI & Automation Engineer
Cluj-Napoca, Romania
Actions
I build production AI agents and intelligent document processing for operational enterprises — banking, payments, government, and finance — from Cluj-Napoca and Timișoara.
Wittenberger Industries ships agentic automation where judgment and unstructured input matter, and RPA where the steps are fixed. Recent work includes hyperscale IDP on Kubernetes (5,000 docs/hour, ~14× faster than a robot fleet), agentic voucher validation (~71% faster, 72h → 18h cycles), a Teams-native ERP assistant (Mia), and an Outlook-embedded sales agent grounded in internal precedent.
Previously at UiPath I led technology services teams on Document Understanding and Human-in-the-Loop. I still speak and build at the intersection of UiPath, Python, agents, and cloud scale.
Site: https://wittenberger.io
LinkedIn: linkedin.com/in/sergiu-wittenberger
GitHub (HyperDU early public cut): github.com/UiPath-Services/HyperDU
Links
Area of Expertise
Topics
Enterprise examples of Agentic Automation
Concrete production stories of agents in the enterprise: voucher processing with Claude + UiPath (~3,500 tx/month, rule-perfect validation, 72h → 18h), an Outlook sales agent grounded in internal precedent, and Mia — a Teams-native agent that replaced Workday/Jira timesheets with chat. Where agents win, where pure RPA still wins, and how we evaluate and guardrail them before they touch real work. In-house, our team's timesheets and HR admin run through TimeKeeper, a chat agent on an MCP connector.
Agents vs RPA: pick the boring tool on purpose
A decision framework from shipping both. Fixed steps + structured inputs + stable UI → pure RPA. Judgment, unstructured docs, or unstable UI → agents. Mixed stacks and how we keep cost, audit, and first-pass accuracy (98%+) under control when you combine them.
Witloop: from conversation to open PR across several AI harnesses.
Most agent coding demos stop at a chat transcript. Witloop is an open plugin that runs a Spec-driven loop from idea to pull request on four hosts: Cursor, Claude Code, GitHub Copilot CLI, and Grok Build. Main steps are: brainstorm, research, plan, build, check, and ship. I'll show how /wit:dev routes feature vs bug-fix vs investigation, and how with one command you go from idea to full implementation, what lands in .wit/ on a project. Repo: https://github.com/Wittenberger-Industries/witloop
A Grok Bot assistant team does my boring jobs
I run a team of Grok Bot assistants from one chat. A front-door bot takes each request and hands it to a specialist. One books padel every week. One keeps timesheets in order. One watches flight prices for family trips. One plans the kids' school-break holidays. Nothing goes out without my approval. I'll show how the routing works, which steps run as plain scripts instead of model calls, and what still breaks.
From a design document to a deployed workflow in a few hours
You give an agent a PDD and together with the witloop skills the agent generates the workflows. A few hours later the process is in production.
My job hunt runs on a bot team: scouting, drafts and applications with a human on the send button
In one week, a team of AI agents worked through 43 job leads and applied to roles on LinkedIn, Freelancermap and company career sites. They wrote 26 tailored pitches, 17 to employers and 9 to integrators. The same team checked 48 tech events across five Romanian cities and Europe, found the ones with open speaker calls, and matched each event to the right talk. Nothing went out without my approval. I'll show the setup. A router bot hands work to specialist bots for lead scouting, pitch drafting, CFP scouting and verification. I'll also show the guardrails. There are hard bans on certain clients and claims, an approval card before every send, and rules learned from my corrections, such as no salaried contracts, a rate floor, and a neutral greeting for general inboxes. You'll leave with the architecture, the failure modes, and the real results: what got replies, what got rejected, and what I had to fix by hand.
TimeKeeper: our timesheets and HR admin run through a chat agent
We gave our small team's management system to AI agents through an MCP connector. Sign-in goes through Microsoft Entra, and there is no website login to fall back on. People log hours and fetch their own certificates and CVs by chatting with the team TimeKeeper bot. A separate admin bot, used only by me, can upload and read anyone's documents. A nightly job copies the documents to SharePoint. I'll show how the permissions are split between the two bots, what broke along the way, and why we built an MCP connector instead of screen automation.
Sergiu Wittenberger
Founder - Lead AI & Automation Engineer
Cluj-Napoca, Romania
Links
Actions
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top