Services
Our company has been building blockchain-based financial projects for over 10 years. Our scope of activity includes the development of centralized and decentralized crypto exchanges, crypto bots, payment gateways, real estate tokenization, DeFi and NFT projects.
AI & ML
Since 2023 we have been building AI for business: custom AI and LLM development, AI agents, chatbots and enterprise AI that connect to your data, products and processes.
Custom AI chatbots that answer from your own data and know when to hand off
Merehead builds custom AI chatbots for customer support, product assistance and internal knowledge. We connect the bot to your help center, product data and business systems, design the conversation, and test answers and handoffs before launch.
Most chatbot projects start with one specific problem. These are the four we design most often.
“Where is my invoice?” or “How do I turn on two-factor login?”
Help center articles and read-only account data through your APIs.
A direct answer with a link to the right article, or a support case opened with the conversation already attached.
Where automation stops: refunds, disputes, account changes and anything that needs a human decision go to an agent.
“How do I set up a recurring payout?”
Product documentation, release notes and the feature list for the user’s plan.
Step-by-step guidance inside the app, with the screen they are on taken into account.
Where automation stops: the bot explains settings but changes them only when the user confirms and the action is on a list you approve.
“What is the procedure for a failed wire transfer?”
Wiki pages, policy documents and templates, searched with the employee’s own access rights.
An answer with the document name and version, so they can check the source.
Where automation stops: if no approved answer exists, the bot says so and sends the question to the document owner.
“Do you support our payment provider? What would a pilot cost?”
Your product catalog, the sales calendar and the CRM.
A few qualifying questions, then a booked call or a new CRM record for the sales team.
Where automation stops: the bot does not quote prices outside the ranges your sales team has approved.
Have a different problem in mind? Tell us which questions your users ask most and which systems hold the answers.
Discuss your use caseThree delivered chatbots, each with set limits, a handoff path to a person and a confirmed result.
Customers kept asking the same questions about workspace access, invoices and account settings. The answers existed, but they were spread across the help center and account records, so support staff searched before they could reply.
An in-app chatbot connected to the help center and to read-only account APIs. It checks workspace access before it shows account details and links to the right instructions.
Billing disputes and out-of-scope requests go to a support agent with the conversation attached.
Four-week pilot: 336 of 600 eligible conversations. The share is of the pilot cohort, not of all support demand.
Deposit questions need the right account, network and transaction record. A general-purpose bot could explain confirmations but could not show a customer the status of their own deposit.
An authenticated chatbot that reads deposit status and transaction records through scoped, read-only APIs, and checks the network and timestamp on each response.
No withdrawal approvals, KYC changes or trading advice. Missing or inconsistent records open a support case with the details already gathered.
Compared with the four weeks before launch.
Employees asked colleagues for procedures they could not find. Several documents covered the same process, and old versions still appeared in search.
A chatbot inside the company workspace. Each answer cites its source and version, and results follow the employee’s role. We reviewed the document set with its owners and removed obsolete material from the index.
Questions without an approved answer go to a named content owner.
Measured in a controlled task exercise.
Have a similar support or knowledge problem? Let’s scope your chatbot.
Discuss your chatbot projectWe assess whether configuring an existing platform, extending it with custom connectors, or building a dedicated chatbot best fits your requirements.
You need answers to a small set of questions with public answers.
The platform covers the conversation, but someone has to build the connection to your CRM, help desk or product APIs.
Available platforms cannot meet your workflow, access, deployment or interface requirements at an acceptable total cost.
We go through each point with you during discovery and recommend the option with the lowest total cost that still meets your requirements.
Whether the platform already reaches your CRM, help desk and product APIs, or whether someone has to build each one.
Whether the bot can show each user only what that user is allowed to see.
How much control you have over the widget, branding and mobile behavior.
Where the system runs, and whether you can move it later.
What you pay per seat, per conversation or per integration as volume grows.
Who holds the prompts, content, conversation logs and integration code.
Who fixes a broken integration on a Friday evening.
Each service ends with a concrete output you can review.
Discuss your projectWe can assess it first. You receive a written review of what works, what is risky to keep, and a fixed-scope plan to make it production-ready.
Request a prototype reviewWe review your support tickets, current help content and the questions people actually ask.
We write the dialogue flows, clarifying questions, follow-up logic, tone and the limits of what the bot answers.
We prepare the content, build retrieval over it and configure the model.
We connect the bot to your website, app or messaging channels and to your systems.
We test against real question categories, missing data, permission checks and handoff cases, then release in stages.
Support is a separate agreement covering monitoring and bug fixes.
Each integration depends on three things:
We confirm all three during discovery, before the schedule is set.
For factual questions, we design the chatbot to retrieve information from approved sources. Account-specific answers use authorized APIs. We test how it handles missing information, conflicting sources and requests outside its scope.
We prepare your documents: remove duplicates, name an owner for each one and set a version.
The bot shows a link to the source with each answer, so the customer or employee can check it.
When you update a document, the bot uses the new version after the index refreshes. You do not retrain the model.
Users see only the content their role allows. A support agent and an internal auditor may get different answers to the same question.
When two sources disagree, the bot says so and flags the conflict for an owner.
When no approved source covers a question, the bot says it does not know and routes the question.
Indexing your documents is not the same as fine-tuning a model on them. It is also not a promise of zero errors. Language models still make mistakes, so we measure how often they happen and design the handoff for the cases that carry risk.
A good bot asks before it guesses. If a question could mean two things, the bot asks which one. It also keeps track of earlier messages, so the user does not repeat account details or a product name in every reply.
Voice is possible, but it is an option and not part of the standard package. We recommend it only where phone calls are already part of the support flow.
The customer does not have to explain the problem again.
We design access around your existing identity system.
The bot knows who the user is before it reveals account data.
Read-only by default. Write actions need an explicit allowlist and, where needed, a confirmation step.
API keys and tokens are stored in your secret manager, not in the prompt or the code.
We agree which conversations are logged, for how long, and who can read them.
The bot cannot run an action you have not approved, even if a user asks.
Access permissions and action limits are enforced by the application and backend services. Testing includes attempts to expose restricted data or trigger unauthorized actions.
Chosen per project, based on data-handling terms and test results.
In the client’s cloud account.
Discuss your requirementsThe application runs in your cloud account, while requests go to the model API you choose. Standard APIs from OpenAI, Anthropic and Google do not use your data for training under their current terms.
We confirm the terms of the chosen provider before launch.
Before a pilot, we agree on the test plan with your team.
Checked by a reviewer.
Whether each answer is backed by an approved document.
Without a repeat contact within the agreed window.
Scored by the operators who receive the cases.
The response time that 95% of answers stay within.
What one conversation costs to run.
An abandoned conversation is not counted as resolved.
Schedules depend on how quickly your team can provide API access, review drafts and clean up source documents. We do not set a date before discovery shows what is involved.
We review your questions, documents and systems.
Your partAccess to sample data and an owner for the content
We build the dialogue flows and test them with your team on sample questions.
Your partYou approve the scope before the build starts
We connect the knowledge sources and systems.
Your partAPI access, credentials and review access
We run the agreed test plan and share the results.
You getTest results, including failures and their causes
The bot goes live for a limited set of question categories.
Your partYour team reviews conversations before the scope grows
We monitor quality, update content and report on the metrics above.
You getRegular reports on the agreed metrics
For the full development process, read How to build a chatbot from scratch for your business
The cost of an AI chatbot has five parts. We show each one separately so you can see what drives the budget.
$3,000–7,000
2–4 weeksA chatbot on your website that answers typical questions, built on an LLM API.
Does not include document upload or a knowledge base built from your files.
$7,000–15,000
4–7 weeksAnswers from documents, PDFs and a knowledge base.
$12,000–30,000
6–12 weeksCRM and API integrations, analytics.
$30,000–80,000
3–6 monthsUser roles, SSO, several data sources, security, audit logs.
Discovery starts at $2,000 and is included in the package price. Each price assumes the channels, integrations, data sources, languages and access complexity in your estimate. Monthly operating costs are estimated separately for cloud infrastructure, model API usage and support, based on your expected monthly conversation volume and average conversation length. We state those assumptions in the estimate.
Some stages run in parallel, so the total is shorter than the sum of the stages. Security approvals and API access are the most common causes of delay.
Get a scoped budget| Stage | Typical duration | What happens |
|---|---|---|
| Discovery and architecture | 3–5 days | Requirements, scenarios, LLM choice |
| UI/UX design | 4–7 days | Chat interface, admin panel |
| Backend and LLM integration | 1–2 weeks | API, context, conversation history |
| RAG implementation | 1–2 weeks | Embeddings, search, documents |
| Cloud deployment | 3–5 days | Docker, CI/CD, database |
| Testing and optimization | 1–2 weeks | Answer quality, security, performance |
Chatbot quality depends on the systems behind it.
Our team has built APIs, permission models and integration layers for fintech and crypto products, so we treat account-linked answers as an engineering task, not a scripted reply.
This is our strongest ground. Account-linked support, transaction status questions and compliance-sensitive answers need different rules than a general FAQ bot, and most of our team has built that kind of system.
Yuri Musienko, CBDO at Merehead, has worked on AI development and optimization, trading platforms and asset tokenization. He has consulted companies on strategic planning and market entry since 2018.
You receive documentation your team can use without us.
Here is what the teams we worked with said after launch.
“We started with access questions and invoice requests because those were taking up time every day. The useful part was the handoff: our team could see what the customer had asked and what the bot had already checked. We also had a clear list of questions it was not ready to answer.”
“The chat interface was only part of the job. We needed account access handled properly, and we needed the bot to say when the transaction data was incomplete. The team worked through those cases with our backend developers before launch. That gave us a much clearer basis for the release decision.”
“The first review exposed how many old versions of our procedures were still in circulation. We sorted that out before rolling out the bot. People can now open the source behind an answer, and each document has an owner. That has been more useful to us than adding another search box.”
Since 2018, Yuri has been consulting companies on strategic planning, entering international markets, and scaling technology businesses. Ask him about your case directly.
Talk to an expertPackages run from $3,000 for a basic chatbot to $30,000–80,000 for an enterprise build. Discovery starts at $2,000, and every estimate lists the assumptions behind it.
A basic chatbot takes 2–4 weeks, and an enterprise build takes 3–6 months. Stage durations are in the cost and timelines section.
The documents you want the bot to answer from, access to the systems it should read, and a list of the questions your team gets most often. Messy content is normal. We help you sort out what is current and what should be removed.
Yes. We start with an assessment and give you a written review of what can stay, what needs rebuilding and what it costs to reach production.
If your CRM offers an API, usually yes. We confirm fields, permissions and rate limits during discovery. Some CRMs need a licensed integration or a custom connector, and we tell you which case applies.
English, which we have tested. Other languages can be scoped for a specific project.
The bot hands the conversation to an operator when the case falls outside its scope, when the user asks for a person, or when the request needs a decision. The operator sees the full history and the data already checked.
The chatbot runs in your cloud account. We hand over the code, documentation and prompts. The model API is licensed separately, under your own account with the provider.
The project includes a 1-month warranty for defects within the agreed scope. Ongoing monitoring and maintenance are covered by a separate support agreement. New integrations, scenarios, model changes and feature improvements are priced separately.
Describe what the chatbot should do. We will review the scope and identify the next steps.