Case studies, competition recaps, and thoughts on AI automation — from building diffusion models at Sberbank to deploying voice agents for UK businesses.
Hackathons, awards, talks & articles
Inside Inbox Router by HermesOps: AI email classification that routes shared inboxes in 1.2 seconds. Plain-language rules, full audit log, on-prem LLM option.
How HermesOps built Inbox Router for two manufacturing clients — 16 inboxes routed in 2 weeks, leads reach sales in 1.2 seconds, everything logged to CRM.
Qwen3.6-27B runs locally on 16-18GB RAM with 262K context and Apache 2.0 license. On-premise LLM automation is now viable for document classification, RAG, and internal chatbots.
Major update to the Lezgin translator: significantly improved quality via synthetic corpus, integrated text-to-speech, and 1000+ volunteers behind the project.
Lessons from selling AI automation services — why understanding buyer psychology is the key to closing deals for AI agencies, SaaS, and consulting.
Featured in a media post by @tellsef covering the journey from school hackathons to Senior AI Engineer — presenting to government leaders, building Kandinsky and X-FLUX, and founding an AI agency.
After phased rollout with ElevenLabs, Twilio, and Vapi, our AI voice agent captured two qualified leads on its first day handling 100% of inbound calls.
Built a Telegram AI assistant for tutors using GPT-4 Vision, LangChain, and RAG — lesson analysis, homework generation, and progress tracking. +10% sales conversion.
How we built an AI voice agent with ElevenLabs and Twilio that replaced a £2,000/month outsourced call center for a UK surveying firm — saving 83% on reception costs.
Docker MCP Toolkit connects Claude Desktop and Claude Code to 312 tools — Notion, Slack, YouTube, CRM, databases. Free, secure, and the best MCP setup for AI agents.
Lezgi Star Award
Honored with the Lezgi Ghed award for building the first Lezgin language AI translator and text-to-speech system — preserving a minority language through technology.
1st place
Won 1st place building a RAG-based resume matching system using LLMs and vector search — the same tech stack behind our AI agent work.
2nd place
Built a multimodal video classification system combining visual and audio features at one of Russia's top hackathons.
Top 5% — Silver Medal
Placed 58th out of 1,000+ teams in Google's Isolated Sign Language Recognition challenge on Kaggle.
1st place — Absolute Winner
Won 1st place at AIIJC, pitched the winning AI solution to government leaders on national television. Built a sign language recognition system using deep learning.