← Yuriy Chukaev · PortfolioTrack record
Ten years · ~25 delivered projects

The full track record

For about ten years I've run a small independent studio, delivering products, integrations and automation to clients end to end — from the first conversation through architecture, code, deployment and years of production support. I owned the whole chain, so I learned to move fast and independently.

Below is the honest long version: what I built, for which kind of business, and the scale it reached in production. Most clients run inside regulated CRM and payment systems, so I describe them by industry and show the systems live on a screen-share.

Core zone · last ~2 years

AI & LLM systems in production

Building and shipping production LLM systems — mainly on Anthropic Claude and Google Gemini — has been the focus of my recent work: speech-to-text pipelines, structured extraction, context retrieval, and deep integration into live CRM/ERP systems. The genuinely agentic engineering — multi-agent orchestration, browser tool-use, verified workflows — lives in the internal ops platform I run my studio on.
Own AI product

AI sales-department controller

Listens to calls and chats, auto-fills the CRM, scores managers and fires alerts on risky or overdue deals. Deployed at three companies across different industries.

85,000+ transcriptions · 69,000+ unique calls · ~1,300h of audio · 31,000+ automated quality checks in 2.5 months — at ~$1/day of compute.
Python · speech-to-text · LLM scoring (Anthropic / Gemini) · CRM autofill
Flower retail chain

Omnichannel AI sales consultant

Reads customer chats across three channels into one CRM window, sees live stock and price, recognizes repeat customers and drafts the order from the conversation. The owner edits how it talks via versioned prompt blocks, no programmer needed.

Trained on 33,000+ real store messages · ~1,000 messages/week · 17 minutes from "hello" to a paid invoice.
LLM consultant · Telegram / VK / Max connectors · amoCRM · inventory API
Open source

Conversation-analysis LLM pipeline

A modular Python platform that exports chat histories, transcribes audio/video, then runs a multi-level LLM pipeline: incremental summarization with content-hash caching, batch-API cost optimization, embeddings and semantic drill-down, plus automatic task/project extraction and dedup.

Includes a two-way chat ↔ task-tracker sync that only invokes the LLM on borderline embedding-similarity matches to control cost.
Python · Anthropic (Claude, batch API) · embeddings · Telethon · SQLite
Auto-repair shop

1C automation via AI

A bot that turns a vehicle-registration photo into client + car records in the accounting system, and a customer's voice/video complaint into a draft work-order — transcription plus an LLM, integrated over the ERP's API.

Grew from a two-hour bonus into a deployed project with ongoing support.
LLM · speech-to-text · 1C OData · Telegram bot
Platforms

E-commerce & marketplace systems

Pharmacy chain · 2 countries

Custom pharmacy e-commerce engine

Migrated a multi-store pharmacy chain (6 legal entities, two countries) off a boxed CMS onto a custom PHP 8 engine without stopping sales: two-way ERP integration, payment routing per entity, fiscalization, delivery, regulated-drug marking, and a second-country storefront with local pricing and acquirer.

Catalog of 35,000+ products (22,000+ AI-classified), 14,000+ live-sellable across 6 regional storefronts; second-country market launched in ~1.5 months.
PHP 8 · MySQL · ERP REST/webhooks · payment gateways · fiscalization · drug-marking APIs
Marketplace fulfillment operator

Warehouse-to-marketplace gateway

A warehouse-to-marketplace gateway with per-parcel scanning and a daily operator UI. Rebuilt v1 (PHP/MySQL) into v2 (FastAPI + React + PostgreSQL + Docker) — 76 functions across 6 domains — and migrated it with eight months of zero-downtime parallel run. Also built a cross-border proxy-API gateway from scratch in one month.

Scaled from ~1,000 to 40,000–50,000 parcels/day (≈40×) over four years · ~1M parcels in the database · team of 5 devs + PM + QA.
FastAPI · React · PostgreSQL · Docker · Flask/Celery/Redis
The bread and butter · ~15 businesses

CRM engineering & business-system integrations

The backbone of the studio: implementing, integrating and running sales-and-operations systems (CRM, telephony, messengers, ERP/1C, billing) for businesses that keep the relationship for years. Named where the client allows it, by industry where it doesn't.
Large logistics company≈5 years
End-to-end CRM engineering for a high-volume operation: telephony, marketplace and messenger integrations, license and process optimization. CRM holds 135,000+ deals and 171,000+ contacts; killed a bot attack (88 spam entries/day from 85 IPs → 0) in two days; restored 7,500+ CRM objects after a bad deletion; built an AI-agent MVP overnight that the client paid for the same day.
Transport companySince 2018
A lead-distribution engine that auto-assigns inbound requests across a 14-manager sales team by configurable rules, plus site-to-CRM intake with UTM and dedup. 43,000+ unique leads distributed over two years, still running daily; zero site-intake errors over 4.5 months.
Wholesale distributor≈5 years
Continuous CRM support for a wholesale business: full sales cycle, automatic loyalty tiers written into the contact card, high-value-segment touchpoints, and a one-day migration of working numbers between messengers. CRM holds 28,000+ contacts and 81,000+ deals.
Dental clinicDr. Gusev
Grew from a CRM setup to a full stack: a Telegram appointment bot, a multi-page site, a mobile scheduling PWA with automatic patient notifications, and the signature win — per-channel messenger attribution that standard UTM can't do, writing each ad source straight into the CRM.
Japanese-car importer≈4 years
Turnkey CRM with two-timezone routing: an 8-800 IVR office selector and duty-manager routing across two regions each on local time, end-to-end analytics, and a CRM ↔ 1C integration for automatic invoicing.
Paint & coatings manufacturerKraskovar
A diagnostics engagement: found and fixed in one day a seven-month silent data-format desync (names, emails and order amounts weren't landing in deals), backfilled the historical deals with zero manager-edit loss, and turned the lesson into a recurring auto-audit of all 66 site-intake schemes.
Second pharmacy chain8 months
Moved a pharmacy chain to end-to-end electronic document flow: 1C ↔ cloud-accounting exchange, drug marking with crypto-tailed scanners, a procurement-platform module, and 1C migration — price lists, orders and invoices flowing without manual work.
Roller-blind manufacturerWholesale + dealers
Reconfigured the CRM for a wholesale/dealer operation in ~3 weeks: a dedicated dealer pipeline isolated from direct requests, deduplication cleanup, and automatic company-details fill from the tax ID.
Trading company≈4 years
CRM implementation and support, messenger channels, telephony, license management and telecom call-marking registration, on a multi-year retainer.
Clothing retailerSince 2024
CRM from scratch with warehouse sync and three messenger channels (WhatsApp, Telegram, Max), plus monthly manager support.
Medical education center≈5 years
CRM audit, automations and CRM-marketing for event enrollment, with the contact base synchronized to an email-mailing service.
Hookah delivery service5+ years
Automated the business since 2021: a post-order NPS survey bot, CRM with a messenger channel, and ongoing support.
Vocational-education centerASO
Connected the site's request form to the CRM so inquiries create deals automatically — the single largest user of my site-to-CRM connector, with 17,000+ requests sent in 1.5 years.
Beverage brandFlavor Splash
A public Telegram giveaway bot (entry intake, receipts, QR codes); later removed the entry cap, migrated it to fault-tolerant hosting, and built a second bot for the follow-on campaign.
Dance studioBee in Dance
An interactive schedule grid on the site with online booking — hall rental and group classes on one screen; still live a year after delivery.
CRM-integrator agency (Uzbekistan)Partnership
Telephony integration (PBX ↔ webhook ↔ speech-synthesis ↔ CRM), then a partnership to automate the agency's own delivery process so it could scale to 200+ implementations per month.
Managed AI-server line2 clients
Provisioned and run managed Claude Code AI servers for two clients as a subscription: dedicated accounts, hardened Linux (firewall, fail2ban, no-password SSH), OAuth auto-refresh, remote-SSH access.
Productized

Own products & tooling

Boxed product · since 2021

Site-to-CRM connector

A productized connector that pipes website form submissions into the CRM with UTM tags and deduplication, sold direct and through integrator partners; rewritten in 2024 on a queue-backed architecture.

341 CRM accounts connected over its life · ~18,000 requests/year · top user sent 17,000+ requests in 1.5 years.
PHP · Vue · Redis queue · CRM API
Widget platform

Self-hosted CRM widget suite

Cloned and self-hosted a suite of ten CRM sub-widgets so the studio owns its own widget/vendor platform instead of depending on a third party — and fixed a query-performance bug in the process.

All ten sub-widgets verified live on a client account · a slow catalog query cut from 250s to 6s.
CRM widget SDK · PHP · JS
Automation infrastructure

Browser-automation bridge

My own browser-automation layer that lets an AI agent drive a real browser for everyday workflows and for reaching web tools that have no API. I run it across a small fleet of persistent identities.

Reused for both data-collection and "prompt → generate → upscale → assemble" media pipelines.
Playwright / CDP · persistent profiles
Engine

Semantic dedup engine

A record-deduplication engine that merges duplicate CRM entities with a semantic arbiter — programmatic rules first, an LLM only for the ambiguous cases — packaged as a reusable module.

Turns a recurring, error-prone manual cleanup into an automated pipeline.
Python · embeddings · LLM arbitration
Foundations · 2012–2019

Earlier career

Independent studio founded2019
Founded the studio that became my full-time work: CRM implementation and integration, then integrations and automation, then the AI/LLM systems above. Everything from 2019 onward is one continuous body of work.
Certified CRM integrator2018
Became a certified CRM integrator and developer and began building the client base the studio still runs on.
Graduate coursework · Software Engineering2017–2018
A year of master's-level Software Engineering coursework at Innopolis University (MSIT-SE program, with Carnegie Mellon University): software architecture, project management and communication.
Freelance & project work≈2014–2017
Web and Android development across freelance projects, including a logistics/registration product and an international medical project — the years I learned to ship independently across whatever stack a project needed.
Full-stack developer · C#/.NET & Angular≈2014–2017
Early commercial web development: C#/.NET (ASP.NET MVC) on the backend and Angular on the front end, across product companies and a delivery/registration project, before moving my primary stack to PHP, Python and JavaScript.
B.Sc. Applied Mathematics & CS2016
Degree in Applied Mathematics and Computer Science (Mathematician-Programmer), Kazan National Research Technological University. First programming (Pascal, algorithms) back in a math-and-language lyceum.