Workflow audit
We map how the work is done today, estimate the hours each step costs, and rank what to automate first by payoff and risk. You keep the document whether or not we continue.
DISCOVERYAI workflows and automation for the work your team repeats.
We find the repetitive work in your business, such as sorting an inbox, copying data between tools, or drafting the same documents, and build AI workflows that do it reliably. A person approves the steps that need judgment, and every run is logged.
- TRIGGERAn invoice arrives by email
- MODELRead the PDF. Extract vendor, amount and due date
- CHECKMatch it against the purchase order
- PERSONApprove anything over $5,000
- ACTIONPost to accounting, file the PDF, tell the owner
- BUILDING SOFTWARE THAT HAS TO STAY UP
- 25+ years
- OF OUR OWN, LIVE AND USED BY THE PUBLIC
- 3 AI products
- LISTS THE LEMONVITE SERVER WE BUILT AND RUN
- MCP Registry
- FROM THE FIRST CALL TO PRODUCTION, WITH NO HANDOFFS
- 1 team
01 / WHO THIS IS FORIF THIS IS YOU, WE SHOULD TALK.
Work that should
not need a person.
Good candidates for automation are frequent, follow rules a person could explain, and have a result someone can check.
01
People copy data between tools.
The same details are typed from email into a spreadsheet, then into the CRM or the accounting system. Each copy costs time and adds errors.
02
A queue is sorted by hand.
Someone reads every support request, lead or application to decide where it goes and how urgent it is, before any real work starts.
03
Documents are read and retyped.
Invoices, contracts, forms and reports arrive as PDFs, and a person extracts the same fields from each one.
04
Your AI pilot never reached production.
The demo looked good, then it gave a wrong answer and nobody trusted it. It needed tests, approval steps and logs, not a better prompt.
02 / WHAT YOU GETAI WORKFLOWS & AUTOMATION
Automation you can
check and trust.
We start with one workflow, put it into production, and measure it. The second one is faster because the foundation is already there.
Document and email processing
Classification, field extraction and routing for invoices, contracts, forms and inboxes, with a confidence score that decides when a person takes a look.
INTELLIGENT DOCUMENT PROCESSINGAssistants on your data
Internal tools that answer questions from your documents, tickets and databases, cite their sources, and respect who is allowed to see what.
RETRIEVAL & KNOWLEDGEAgents and integrations
Agents that take actions in your systems through their APIs, and MCP servers that let Claude or ChatGPT work with your product safely.
AI AGENTS · MCPDrafting and reporting
First drafts of replies, proposals, summaries and weekly reports, written from your data in your format, ready for a person to edit and send.
GENERATIONReliability built in
Evaluation suites that catch regressions, approval steps for risky actions, a log of every run, retries that resume after a failure, and a known cost per run.
EVALS · MONITORING03 / WHO LEADS THE WORK RAN MAGEN · FOUNDER
Built by engineers,
not a prompt.
A workflow that runs your business has the same requirements as any production system: it must handle bad input, recover from failures, and show what it did. I’m Ran Magen, the founder, and building systems like that has been my job for 25 years.
Altys
CO-FOUNDERAn AI learning app for kids. I run the technical side of the company: the Go backend, the cloud infrastructure, and how we ship.
Lemonberry Labs
FOUNDER & CEOI build, launch and support products with paying customers, Lemonvite among them, and I lead every client project. It taught me what engineering work a business needs.
Coinbase
TECH LEAD, BACKEND PRODUCTIVITYI was the technical lead for backend engineering productivity. I led the migration to GraphQL Federation, set Protobuf standards across more than 700 repositories, and founded the Server Foundations Guild.
Apollo GraphQL
ENGINEER, FEDERATION & INFRASTRUCTUREI worked on backend infrastructure and Federation, and wrote the next-generation Federation gateway in Rust.
Amazon Web Services
ENGINEER, MEDIACONNECTI built AWS Elemental MediaConnect, a live video transport service, from its first line of code.
I was the tech lead for Timelines infrastructure: Home, Profile and Notifications. I owned tweet fanout, the system that writes each new tweet into its followers’ timelines, for hundreds of millions of users.
Startups
ENGINEERI built distributed systems at Dapper (acquired by Yahoo), my6sense, and Worklight (acquired by IBM), after eight years building for the web.
04 / PROOFBUILT, SHIPPED, OUT IN THE WORLD.
AI we already
run in production.
We use the same techniques in products of our own, where real people depend on the result.
01
Lemonvite
MCP SERVER · AGENT INTEGRATIONAn MCP server that lets Claude and Codex create, send and track invitations. Listed in the official MCP Registry.
02howtowritea.com
DOCUMENT GENERATIONTurns a few answers into a complete demand letter in minutes.
03AutoDoggy
PERSONALIZED GENERATIONGenerates a dog training plan from each owner’s answers about their dog.
04duro
OPEN SOURCE · DURABLE WORKFLOWSA Go library for pipelines that checkpoint every stage to Postgres and resume where they stopped after a crash.
05Cloud Workflows emulator
OPEN SOURCE · WORKFLOW TOOLINGA local emulator for Google Cloud Workflows, so workflows can be tested before they are deployed.
05 / HOW IT WORKSFROM FIRST EMAIL TO DONE.
One workflow first.
Then the next.
- 01 /
Demo call
You describe a process. We show you a comparable workflow running, and we work out what yours would take.
- 02 /
Audit and proposal
We map the process with the people who do it, then send a written proposal: what gets automated, what stays with a person, the expected saving, and the price.
- 03 /
Build and test on your data
We build the workflow against real examples from your business and measure how often it is right before it touches live work.
- 04 /
Launch, monitor, extend
It goes live with approval steps and logging. Once the numbers hold, we reduce the approvals and move to the next workflow.
06 / QUESTIONSASKED BEFORE MOST FIRST CALLS.
Questions,
answered.
Something else? Write to ran@lemonberrylabs.com.
What kinds of work can AI automate?
Work that is frequent, follows rules a person could explain, and has a result someone can check: sorting and answering email, extracting data from documents, moving information between systems, drafting replies and reports, and answering questions from internal knowledge. Work that depends on relationships or on judgment nobody can describe stays with people.
Which tools and AI models do you work with?
Models from Anthropic, OpenAI and Google, chosen per task by accuracy and cost. We connect them to the tools you already use through their APIs, including email, spreadsheets, CRMs, accounting systems, help desks and internal databases.
Is our data safe?
Workflows run in your own cloud and vendor accounts, with access limited to what each step needs. We use the providers' API offerings, which by default do not train on your data, and we document where every piece of data goes.
What happens when the AI gets something wrong?
We design for it. Each step has a confidence check, risky actions wait for a person's approval, every run is logged so mistakes can be traced, and an evaluation suite built from your real examples catches regressions before a change goes live.
How long does it take?
We start with a single workflow so you see a result in production quickly. The timeline depends on how many systems it touches and how clean the data is, and the proposal gives you a date.
How much does it cost?
The audit and each workflow are priced separately in a written proposal, alongside an estimate of the hours it saves. Running costs for models and hosting are estimated up front and monitored after launch.
Do we need engineers on our side?
No. We build, deploy and maintain the workflows. We need time from the people who do the work today, because they know the rules and the exceptions.
Software development
Backend systems, infrastructure, web front ends, AI integrations, Mac apps and mobile apps.
ALSO FROM THE STUDIOCoaching
Technical leadership coaching and interview preparation for senior, staff and principal engineers.
07 / NEXT STEP
Show us the work
you repeat.
Describe one process and we will show you what automating it looks like.
Or write to ran@lemonberrylabs.com.