We learn the process, connect to the systems already in place, and take it through production.
Book an intro callThe day-to-day never stops, and pulling processes apart takes hours nobody has.
Nobody inside the organization holds AI adoption and drives it forward.
Hard to track what shifts every month and judge what is actually relevant.
Even with an owner, a secure system built to scale is a project in itself.
We come in and study how the work is actually done. Sessions are recorded and transcribed, so the time your people spend teaching us stays minimal.
We design around the resources you actually have and focus the effort where the impact is highest. Technology choices follow the need and the systems you already run, not the hype of the moment.
A secure system built for scale from the start, on infrastructure that holds as usage grows. Not a POC that falls apart once someone starts working with it seriously.
We go live with real users, with human oversight on every action and corrections along the way, until the system becomes part of how people work.
An accounting firm spends a significant part of every month collecting materials from its clients: invoices that never arrived, missing documents, questions that repeat. It consumes hours of expensive professionals and produces no professional value. We built an AI employee that does it.
We shadowed the person doing the work and recorded her screen through real working sessions, rather than interviewing her about the process. That is how the gaps that never come up in a meeting surfaced: where the client gets stuck, what gets sent twice, what actually consumes the time.
We separated what a machine does well - follow-up, reminders, document identification, recurring answers - from what has to stay with the accountant. The firm sees every conversation and can step into it at any moment. Without that oversight layer, no firm puts a system like this in front of its own clients.
A multi-tenant system from day one: a workflow engine that carries follow-up lasting weeks without dropping it, an agent layer for conversation and document understanding, and integrations into the systems the firm already runs. Not a prototype retrofitted into a product.
A gradual rollout on real cases with a partner firm, with a dashboard where every message is visible and can be stopped, and corrections made alongside the people actually using it.
Temporal and n8n for processes that run for weeks, with retries, saved state and recovery after failure
LLM agents for conversation and decisions, RAG over the client's history, data extraction from scanned documents
WhatsApp Business API, email, and Microsoft 365 / SharePoint for file management
Reading and writing against legacy ERP systems never designed for it
Data separation between clients, permission management, and a dashboard where every action can be stopped
Over 10 years in product management, six of them as Director of Product at Yotpo. Founded a profitable B2B SaaS company in the Shopify ecosystem, where he built and still runs an AI platform that manages paid media across client accounts daily. Technion graduate.
LinkedIn ↗Lead developer. 15 years of engineering experience, including CTO roles at funded startups. Builds fast without trading away the quality of the code or the infrastructure underneath it.
LinkedIn ↗The processes we do our best work on usually have four things in common.
Same shape, different content, over and over. Volume is what makes the build worth paying for.
Someone is chasing someone else for a file, an answer, or an approval, in free-form human language.
An inbox, a spreadsheet, and a core system from 2004 that has no real API and is not going anywhere.
Most of it is routine, but the edge cases carry real consequences and have to stay with a person.
When we say no: one-off analyses, a process nobody can describe because it changes every time, and anything where the honest fix is a better form rather than an AI system.
A workflow that waits, retries, remembers what is still missing, and resumes after a failure instead of starting over.
Reading and writing against core systems that predate the cloud, without asking the organization to replace them.
Scans, photos, forwarded threads and half-answers turned into structured data the rest of the system can act on.
Every action visible, stoppable and attributable, so the organization can hand over work without handing over control.
Multi-tenant separation, security and monitoring designed in at the start, because retrofitting them is a rebuild.
Every engagement starts the same way, and it is free. You walk us through the process as it works today. We come back in writing: what is worth automating, what is not, and what it would take to build.