AVAILABLE FOR SELECT PROJECTS

I build AI systems
that do real work.

I help businesses reduce manual work, handle customers and data faster, and turn ideas into practical digital systems.

I build automation, AI assistants, intelligent document search and complete online products — from an idea to a working system.

From business problem → architecture → working product.

  • Next.js
  • FastAPI
  • Python
  • TypeScript
  • PostgreSQL
  • RAG
  • LLM Agents
01Selected work
SELECTED WORK

Systems built to solve
real problems.

A selection of AI, RAG and backend systems focused on practical engineering rather than demos.

Applied AI / ML01

LLM Hallucination Detector

Problem / task

A lightweight system for detecting factual hallucinations in LLM responses, built for a Sber AI hackathon.

What was built

Evaluates hallucination probability from prompt + model answer, with optional reference answers. Combines structural and semantic features with fast production-oriented inference.

What this demonstrates

Shows a production-oriented approach to evaluating LLM answer reliability with measurable model quality and inference speed.
PR-AUC 0.9076Preview dataset
F1 0.8400Preview dataset
23–38 ms warm inferencePreview dataset
  • Python
  • Sentence Transformers
  • Logistic Regression
  • Docker
  • ML
RAG / Knowledge Systems02

RAG Simble FAISS Consultant

Problem / task

RAG consultant for working with document collections using FAISS and LangChain, including vector-base merging, hybrid retrieval and local document processing.

What this demonstrates

Shows how document collections can be turned into a searchable knowledge workflow with local retrieval and vector search.
  • Python
  • LangChain
  • FAISS
  • RAG
  • RetrievalQA
Multimodal AI03

Multimodal Production Analysis

Problem / task

Multimodal analysis of a confectionery production environment using AI to identify operational bottlenecks, HACCP-related risks and optimization opportunities.

What this demonstrates

Shows how visual information and process context can be combined to surface operational risks and improvement opportunities.
  • GPT
  • Multimodal AI
  • Production Analysis
  • HACCP
Backend / Automation04

Weighted Lead Distribution CRM

Problem / task

Lightweight FastAPI miniCRM with automatic lead distribution based on configurable weights and operator workload limits.

What this demonstrates

Shows backend business logic that routes incoming leads using configurable rules and operator workload constraints.
  • FastAPI
  • Python
  • Backend
  • Lead Routing
  • API
WHAT I BUILD

What I can build
for your business.

From AI automation to complete software products — systems designed around real business processes.

SERVICE01

Routine work automation

AI AUTOMATION

The system can process incoming requests, move data between services, assist customers, work with documents and pass tasks to employees.

  • Customer requests
  • Documents
  • Support
  • Repetitive operations
  • LLM
  • APIs
  • FastAPI
  • Webhooks
SERVICE02

AI assistants for business

AI AGENTS

More than a chatbot. An AI assistant can understand a task, find information, work with your services and complete multi-step actions.

  • Sales
  • Information search
  • Internal assistant
  • CRM workflows
  • LLM Agents
  • LangGraph
  • Tools
  • APIs
SERVICE03

AI that knows your documents

RAG / KNOWLEDGE SYSTEMS

Connect instructions, catalogs, contracts or a knowledge base, and AI can answer questions using your own materials instead of relying only on general model knowledge.

  • Documents
  • Instructions
  • Catalogs
  • Knowledge base
  • RAG
  • Embeddings
  • FAISS
  • Vector Search
SERVICE04

Online services and workspaces

SAAS / WEB PRODUCTS

I build complete online systems: customer accounts, dashboards, internal employee tools and services for clients.

  • Customer accounts
  • Dashboards
  • Internal systems
  • Online services
  • Next.js
  • TypeScript
  • FastAPI
  • PostgreSQL
SERVICE05

Connecting your business services

BACKEND & INTEGRATIONS

I build the internal logic that lets websites, CRM systems, databases, messengers and external platforms exchange data automatically.

  • CRM
  • Messengers
  • Databases
  • External services
  • Python
  • FastAPI
  • PostgreSQL
  • Redis
SERVICE06

AI that understands more than text

MULTIMODAL AI

The system can work with images, video, voice and text together — for example to analyze operations, documents, visual information or conversations.

  • Photos and images
  • Video
  • Voice
  • Process analysis
  • Vision
  • STT/TTS
  • LLM
  • Multimodal AI
HOW I WORK

From a messy process
to a working system.

I start with how the work actually happens today, then choose the smallest reliable solution that removes the problem without adding unnecessary technology.

01

Understand the process

I map what happens now, where time is lost, what gets copied manually, and which decisions really matter.

02

Choose the first working version

I avoid complexity for its own sake and define the smallest reliable system that can solve the problem.

03

Build and connect

Automation, AI, software, data and integrations are used only where they create practical value.

04

Verify real operation

Before launch I check errors, exceptions, security, mobile behavior, monitoring and a rollback path.

WORKING PRINCIPLES

Business process before AI hype.

Critical actions use trusted data, rules or human confirmation.

Security and access are considered before production.

Systems can start small and grow.

Testing, monitoring and rollback are part of delivery.

TRY AI RIGHT NOW

Describe what you want to improve or create.
See which digital solution fits.

In plain language, describe a manual process, problem, website, service, CRM, integration, AI assistant or another digital business task. You will get a concise recommendation without unnecessary complexity.

0 / 1500

Do not enter passwords, card numbers, identity documents, patient medical details or other unnecessary sensitive personal data.

SECURE · BUSINESS ONLY

No technical terms needed — describe the task the way you would explain it to a colleague.

This is a preliminary digital-solution review, not a final specification. Exact requirements and integrations are verified before implementation.

BUSINESS DIGITAL REVIEW
TASK
DIRECTION
NEXT STEP
Ready for your task
FAQ

You do not need to know
what technology you need.

I do not know what exactly to automate. Where do we start?

Describe what people currently do manually, what gets repeated, where information is lost, or what is hard to control. I can help turn that into a concrete first scope.

Does every solution need AI?

No. If normal automation, rules, an integration or a simple system is safer and more efficient, that is what should be used. AI is added only when it genuinely helps.

Can you connect the services we already use?

Often yes, when the service provides a safe supported way to connect. If it does not, I will not pretend an integration exists; we choose another reliable workflow.

Can we start with only one process?

Yes. A focused first version is often the best way to prove the workflow, reduce risk and then expand only where it makes sense.

What about payments and other important data?

Important facts should come from a confirmed source, deterministic rule or authorized person. AI should not guess whether a payment happened or approve critical actions.

What do you need from me to begin?

A plain-language description of how the process works today, who is involved, what is done manually and what result you want. Technical terminology is not required.

Can we work remotely?

Yes. I am based in South Korea and work remotely with projects where the process, access and communication can be organized safely.

ABOUT

I don't just add AI features.
I design working systems.

I'm Eduard Sarmanov. I design and build digital systems that help businesses automate work, use AI and connect different services into one practical workflow.

My specialization is AI Systems Architecture and Full-Stack Engineering. That means I can design the complete system — from what users see to the internal logic, AI, data and integrations.

I'm especially interested in problems where simply connecting ChatGPT or building a bot is not enough. I first understand how the business actually works, then design the solution and turn it into a product people can use every day.

HOW I THINK

Business problem → system architecture → implementation → measurable working result.

CONTACT

Have a problem to solve?
Let's talk.

Tell me what is currently manual, repetitive or difficult to control. You do not need to know what technology is required.

WHAT HAPPENS AFTER YOU MESSAGE

  1. 01You describe the process in ordinary words.
  2. 02I clarify only what is necessary.
  3. 03We discuss a concrete first working scope.