We build automations, applications and AI agents, connect company knowledge and train teams. During a free call, we explore how to improve a selected process. We are based in Warsaw and work with businesses across Poland.
Selected client work
Natu.CareAI collaborationVisit the company
Signature EstatesAI agent for real-estate acquisitionView the case
jakotako.designClaude training and an agent loop for sales-prospect sourcingView the case
foap.comCreator email operations and reply reviewView the case studyWe set the pilot target before the build. A pilot on agreed data usually takes 3–6 weeks. Before it starts, the contract states the target, the fixed price, and how we will check the result. After the pilot, you get the results and decide whether to proceed with the full implementation.
30 min · €0
We discuss one recurring task: inbox handling, invoices, quotes or sales contacts. We learn how it works today and where the difficulties are. The engineering call takes 30 minutes. Within 2 business days, you receive a written summary: what the process costs you per year, what to automate first, from what price, and what not to build.
After reviewing data and requirements
We review sample data, integrations and exceptions. Once we have enough information, we prepare a fixed-price proposal. For a larger scope, we propose an audit from €600. We credit its fee toward the build.
fixed price
Before work starts, the contract defines the scope, data, fixed price and one pilot target: handling time, cases completed without manual work or data accuracy. We agree on measurement, acceptance conditions and the fixes included in the price. You know how we will assess whether the system works.
build time depends on the format · pilot usually 3–6 weeks
The system handles an agreed part of the process using your data, usually for 3–6 weeks. It gets the access needed for the task, records its actions and routes exceptions to a named person. We check the result against the contract target. If it falls short, we complete the agreed fixes within the pilot price. Further changes are quoted separately.
Yours to keep
You get concrete results: what the system achieved, what the full implementation will cost and what work remains with your team. You then decide whether to continue. Code, model instructions, test data and documentation belong to you. You can keep developing the system with us or hand it to another provider. Maintenance scope and price are agreed before the contract.
Deployed
The result guarantee applies to automation, app, and agent implementations. It requires access to the agreed data, enough cases for a reliable assessment, and the scope agreed before the pilot to remain unchanged. Training is priced and accepted under separate terms.
During the 30-minute call, we discuss the problem and how your team works. Within 2 business days, we send a written summary: the process cost calculated on your numbers, what to automate first and from what price, and one concrete next step.
Process scan takeaway
example layout
What it costs today
Copying PDF invoices into the ERP takes 20 hours a month. At a labour cost of €15 an hour, that is €3,600 a year.
What to automate first
Reading invoices and preparing postings for the accountant to approve. Automation from €1,750 net.
Next step
We review sample invoices and the ERP integration, then send a fixed-price proposal.
If a simple script or a later start is the better route, the recommendation appears in the summary.
You receive a fixed price and timeline in a free proposal after we review the data and integrations. For a larger scope, we propose an audit from €600 net, credited toward the build.
Tell us where the workflow stalls and what the team handles manually. We can then scope the right work: connecting tools, building an application or training the team.
When the team copies data between tools and tracks each step manually. We connect the tools you already use and build systems that carry out work across them. This can range from a document workflow to an agent handling a case through several decisions.
When employees search across scattered information or existing tools do not fit the workflow. We build tools for teams and customers: portals, internal workspaces and company knowledge systems with AI. We can also replace selected features of expensive subscriptions with software you own.
When you need to assess a model on company data or run it in your own environment. We select models for the task, fine-tune them on company examples and deploy them in an agreed environment. First, we check whether an existing model with suitable instructions and access to knowledge is sufficient.
When employees need to practise using AI on their own tasks. We teach employees to use AI in their work and technical teams to build with agents. The programme reflects participants’ roles, tools and experience.
Each description explains the task the system handles, when a person decides, and how the work is recorded.
The foap.com interface brings together creator correspondence, campaign context and reply review. Cases requiring a decision go to a person. The case shows current local views with fictional data.
See the deployment
foap.com · local application in development · fictional data
According to Statistics Poland (GUS), 12.5% of enterprises in the Mazowieckie voivodeship used AI in 2025. That is the highest share in Poland, against a national average of 8.7%. Mazowieckie also has the largest share of firms with high or very high digital intensity (39.0%). The survey covers firms with at least 10 employees. GUS does not publish a separate figure for Warsaw itself.
The percentages come from the GUS 2025 survey and cover Poland as a whole. The example tasks describe a possible scope of work. They are not a list of Syntalith clients or projects in Warsaw.
Agents that handle tickets, review code changes, write technical documentation and support a team working with AI coding tools.
A person approves code changes and production releases. The agent works within permissions agreed before launch.
Reading and comparing documents, draft case summaries and search across company templates. A company Claude or ChatGPT account with working rules is often enough to start.
A specialist gives the legal or tax opinion. The system shows the source of every finding.
Sorting a shared inbox, drafting replies and reporting from the CRM.
Messages to clients and data changes wait for staff approval until the pilot confirms quality.
Order status answers, returns and complaints, and reconciling data between the shop and the ERP.
Refunds, exceptions and disputed cases are approved by staff.
Delivery and technical review
More than 22 years of software engineering experience combined. Artem and Serhii own the architecture and delivery. On complex projects, Przemek and Piotr independently review the architecture and stand behind their assessment.
Meet the full team
Co-founder · Applied AI/ML Architect
Has built production systems for more than 6 years. His AI work spans chatbots, model training, fine-tuning, and private deployments through RAG and workflow orchestration to agents that use governed sources and tools. He has worked with AWS Bedrock, OpenSearch, and pgvector.
Co-founder · Voice AI & Agent Systems Engineer
More than 7 years of commercial .NET experience. At Syntalith, he builds voice AI and agent systems: telephony, integrations, evaluation sets, scenario and edge-case testing, and execution-trace analysis.
GenAI Tech Lead · external auditor
More than 5 years in IT, including more than 2 years of hands-on AI and LLM work. He designs RAG and multi-agent systems, runs models on-prem, and leads fine-tuning and evaluation.
Backend & AI Engineer · external auditor
Four years of production Python backend experience. He built APIs, RAG platforms, asynchronous data pipelines, and automated local vLLM deployments on Kubernetes.
Questions before you buy
Times are shown in your own time zone. We work with clients across time zones.
The scan chooses one process to assess, and within 2 business days you receive a recommendation, including when a simpler route is the better fit.