Cortex IT cases
Client projects in production, and our own tools
Client work: a RAG assistant inside Bitrix24, AI scoring for sales calls, computer vision on dairy farms, contract review and counterparty checks. Below those, our own products and open-source repositories. Every case lists the stack, the technical decisions and the go-live date.
Client projects
- Woodworking and metalworking machinery distributor
Machinery distributor: a RAG assistant inside Bitrix24
A RAG assistant in Bitrix24 over 11,000 technical documents. Live since 05.2026.
Read the case
- B2B sales, Bitrix24
Call analytics in Bitrix24: AI scoring for 100% of sales conversations
AI scoring for every sales call against the company's own methodology: transcription, a 0-10 score, results posted to the Bitrix24 deal card. Recordings never leave the client's perimeter.
Read the case
- Dairy cattle farm
Dairy farm: identifying cows from photos and video
A CV system that identifies cows from a photo (mobile app, 78%) and from a live video feed (Re-ID).
Read the case
- Dairy cattle farm
Dairy farm: automatic detection of key events, 24/7
A CV stack that detects key behavioural events automatically. 4 cameras, bird's-eye view, infrared at night. Deployed at the edge on Jetson, no cloud.
Read the case
- Service for brokers and developers
Real-estate agency: an AI assistant for property search and objection handling
A Telegram bot for agents: plain-language property search, a client profile that works like a CRM, and ready-made sales arguments.
Read the case
- Legal function, contract work
AI contract review: diffing against a reference template and flagging risky wording
First-pass contract review: key fields extracted, clause-by-clause diff against a reference template, risky wording flagged with an explanation. From 1-2 hours of reading to a report in minutes.
Read the case
- Due diligence, counterparty screening
Counterparty check by tax ID: an AI agent compiles a report from registries and sanctions lists
A tax ID goes in, a structured counterparty report comes out — sourced, linked, and rated green, yellow or red. From hours of manual searching to a report in minutes.
Read the case
- Automation and AI integration contractor
AI scoring for tenders and leads: from 80-120 postings a day to 3-5 in Telegram
Monitoring procurement and project boards across three sources: postings parsed, the brief scored 0-10 by an LLM, HOT and WARM sent to Telegram. Monitoring used to take 2-3 hours a day.
Read the case
- Rentals and accommodation, several independent platforms
Availability calendar sync: integrating with channels that offer no API
Channels without an API: booking notification emails are parsed and merged into one Google Calendar. Double-booking protection where no integration officially exists.
Read the case
Our own products and open source
Tools Cortex IT built to run its own processes, plus repositories you can read line by line — something client projects never allow.
- Web development and infrastructureCortex IT internal product
cortexit.ru: from content strategy to our own deployment, no site builders
The company site on our own code and our own server: content strategy, design system, front end, Schema.org markup for AI search engines, deployed to an Ubuntu VPS.
Read the case
- Computer vision, segmentationOpen source
Floor plan vectorisation: a raster image into structured walls and openings
A segmentation model turns a scanned floor plan into vector data — polygons for walls and openings. Source code is open and covered by tests.
Read the case
- Backend, request distributionOpen source
Automatic ticket routing: a mini-CRM with operator competence weights
The service distributes incoming tickets across operators by competence weights and current load, with no manual assignment. Source code is open and covered by tests.
Read the case
- Data processing, classification without an LLMOpen source
Job title classifier: matching noisy 1C exports against a closed reference list
A deterministic pipeline matches noisy job titles from a 1C export against a closed classifier — no LLM, no embeddings, no external APIs. Source code is open and covered by tests.
Read the case