Espise builds and runs production AI for legal and commerce teams in the US, Europe and the Gulf: document AI, case intake and customer support that read from and write to your existing tools.
The engineer you talk to is the engineer who builds it
Your data never trains a model
New claimant packetA 186-page scanned PDF arrives from intake.
read 186 pages, 4 handwritten forms
extract injury date, 7 providers, 23 visits
match 7 of 7 providers found in national registry
write case note and timeline saved to case system
To paralegalReady for review. One provider address needs confirmation.
Customer on WhatsAppWill the 65W charger fit my ThinkPad T14 Gen 3? And where is order 48213?
product ThinkPad T14 Gen 3 identified
catalog 2 compatible chargers in stock
order 48213 shipped, tracking found
check answer grounded in catalog and order data
Agent replyYes, two 65W chargers fit your T14 Gen 3, links below. Order 48213 is in transit and due tomorrow.
Illustrative runs. Every step is logged, and anything the agent is unsure about goes to a person.
Shipped work
Our clients' names stay private. Below are the systems, numbers taken from version control, and what changed, described generically. References are available after a first call.
Legal-tech companyUnited States, 2025 to now
Document AI and case analysis for litigation teams
Core engineering on a platform that turns large case files into searchable facts, medical timelines and grounded chat. We built document pipelines, integrations with the document and practice-management systems law firms already use, SOC 2 hardening, and the monitoring that keeps it running.
19 monthsin production engineering
220pull requests merged in six months
2,500+commits shipped
Online retailer2026
Taking over and securing a legacy commerce platform
We took over a legacy .NET platform spread across five servers, documented it and kept it shipping. When a scraping attack hit the product catalog, we traced it and automated defences at the network edge, then connected the platform to its suppliers' stock and price feeds.
60Mrequests in the attack we traced
650+commits in the first seven weeks
~24supplier integrations
Online retailer2026
A support agent designed from real conversations
Before writing agent code, we read the store's support history to learn what customers actually ask and where the answers live. The result is a WhatsApp agent that checks the catalog and order system before replying, and an operator panel where staff can take over any conversation.
2,309conversations analysed
42,385messages reviewed
Plaintiff law firmUnited States, 2026
Case intake automated inside the firm's case-management system
New cases were pulled from the firm's case-management system, key facts were extracted with a mix of language models and rules, treating providers were matched against the national provider registry, and structured case notes were written back for staff to review.
5case types covered
63automated test files
2external data sources matched
Three ways to work with us
Every engagement has a fixed scope, a written outcome and a price you know before we start.
AI systems audit
Fixed fee, 1 to 2 weeks
We review your data, workflows and existing AI or automation, then hand over a written plan: which use cases are worth building, what they will cost, and what could go wrong.
from $3,000
Production pilot
Fixed scope, 4 to 6 weeks
One agent or pipeline built on your stack, connected to your real systems and measured against an outcome we agree on in week one. It ends with a go or no-go decision.
from $8,000
Engineering partner
Monthly, 30 days' notice to stop
A fractional AI lead plus engineers who own architecture, delivery, code review and on-call for your AI features, with a monthly report on what the systems handled.
from $5,000per month
Invoiced in USD or EUR. The final price depends on scope and is fixed in writing before work starts.
How a pilot runs
We plan a typical first project at six weeks from first call to production.
Week 1
Discovery
We read your real data and workflows and agree on one measurable outcome.
Weeks 2 to 5
Build
A working system on your stack, demoed to your team every week.
Week 6
Production
Launch with monitoring, a human handoff path and written runbooks.
After launch
Run and improve
A monthly report on volume handled, accuracy and time saved.
Your data, and who is accountable
The questions your security team will ask, answered before they ask.
No training on your data
We use model providers only under API and enterprise terms that exclude training on customer data, and we tell you which providers each system uses.
Your cloud, your keys
Where possible, systems run in your cloud account with your API keys. We get least-privilege access, and it is removed when the engagement ends.
Retention and deletion
We agree in writing how long prompts, logs and vector indexes are kept, and delete them, embeddings included, when the engagement ends.
Contracts
An NDA before discovery. A data processing agreement on request, written for GDPR and Kosovo's data protection law (Law No. 06/L-082).
Tested before it ships
Every agent is checked against a set of real, held-out cases before launch and after each change, so accuracy is measured, not assumed.
A person stays in the loop
Agents hand uncertain cases to your staff, every action is logged, and we are on call for the systems we run.
Who you work with
Samet Aydın
Founder, engineering lead
Eleven years building software end to end, the last five in AI. First engineer at Truve, a US legal-analytics startup, helping grow it from five to thirty-five people. Previously CTO at Building.tech, where he led an AI system that reads multi-page construction drawings, and at Varsapp, a consumer marketplace.
Tell us about the workflow that eats your team's week.
We reply within one business day. If it is a fit, the next step is a 30-minute scoping call. If it is not, we will say so and point you somewhere better.
Based in Prishtina, Kosovo, on Central European Time. Our mornings overlap with the US East Coast, our afternoons with the Gulf.