An AI bridge on Telegram that runs day-to-day operations for a US service business, linking the owner to an LLM agent and to their live lead flow.

What I built

• Real-time lead capture: Facebook lead ads to Make to Telegram, every enquiry delivered instantly with full context
• A Claude-powered agent the owner talks to in plain language to trigger tasks, drafts and updates
• An n8n pipeline that turns prompts into SEO blog drafts for review
• Group-aware command routing, deferred jobs and status handling so nothing gets dropped

Impact

• The owner runs marketing and content from one Telegram chat, no dashboards
• Leads reach the right person in seconds

Stack: Claude/Anthropic API, Telegram Bot API, Make, n8n, Node.js, Docker.

(Private client system)

A Telegram bot that lets warehouse staff log stock in plain language (typed or voice) and turns it into clean structured inventory.

What I built

• Natural-language intake: staff describe parts received, an LLM extracts product, quantity and attributes into structured records
• A strict trust boundary: regex on the original message is the source of truth for part numbers and quantities, the LLM only adds, never overwrites
• A smart helper that flags ambiguous entries and asks a clarifying question instead of guessing
• Direct writes to the live inventory ledger, with backups and dry-run safety before every commit

Impact

• Non-technical staff update inventory from their phone, no forms or training
• Verbatim guards stop the model from corrupting part numbers

Stack: LLM (Claude/Kimi), Telegram Bot API, Python, speech-to-text, Docker.

(Private client system)

A dispatcher that pushes leads and website analytics straight into a client’s marketing Telegram group, so the team sees what matters without logging into anything.

What I built

• Real-time lead delivery into the marketing group as enquiries arrive
• A daily analytics digest per site: carts, sessions, orders and key activity, posted automatically each morning
• Data pulled from the live sites and analytics, formatted into a clean readable summary
• Multi-site support, each client group gets its own feed

Impact

• Marketing teams get a daily pulse of the store in one message, no dashboards or exports
• Leads and drop-offs surface immediately, where the team already works

Stack: WordPress/WooCommerce, PHP, Telegram Bot API, GA4/analytics, cron, Docker.

(Private client system)

Multilingual WooCommerce store for a kids’ furniture brand (Moldova).

What I built

• WooCommerce storefront with RO/RU multilingual (WPML), redesigned catalog and product pages
• Front-end performance work on the hero/LCP, plus a 3-layer cache setup (OPcache, Nginx FastCGI, Cloudflare)
• MAIB card payment integration and localized order emails

Automation

• Abandoned-cart recovery in n8n: a timed follow-up email with a coupon after checkout drop-off
• Telegram bot giving the owner a daily digest of carts, stock and activity

Stack: WooCommerce, WordPress, PHP, WPML, n8n, Telegram API, Cloudflare, MAIB.

Live: https://www.twine.net/signin

WordPress site and marketing automation for a US garage-door service company (Chicago area).

What I built

• Custom WordPress/Elementor site with LocalBusiness and FAQ schema, click-to-call CTAs, and Core Web Vitals tuning
• Real-time lead routing: Facebook lead ads to Make to Telegram, so every enquiry reaches the owner in seconds with full context
• Automated SEO content pipeline that drafts and publishes local blog posts on a schedule

Impact

• Leads that used to slip through now land instantly, with source and contact details attached
• Organic content keeps shipping with no manual work

Stack: WordPress, Elementor, PHP, Make, Telegram API, LLM content pipeline, schema/SEO.

Live: https://www.twine.net/signin