后端开发者

Company profile: Legal AI platform for the MENA market: collection, structuring and search of legislation and case law. Led two divisions: DATA — collection and processing of legal data, and Backend — platform and API. Team: 13 people. DATA: – Data collection: developed a system for live collection of legislation and case law from more than 34 sources. – Legal instrument processing pipeline: designed and implemented an in-house solution, since off-the-shelf tools could not cope with Arabic and the complex structure of Arabic laws. Having immersed myself deeply in the subject domain, I developed a universal model for stor-ing and structuring legal instruments and achieved high text and structure extraction accuracy of up to 98.7% when compared against reference annotations across the entire data corpus, and also achieved high processing speed of about 0.13–0.2 seconds per page. Processed about 15,000 legal instruments and about 347,000 court cases. Result: a corpus of ~15,000 legal instruments and ~347,000 court cases from 34+ sources in live mode; a 300-page instrument is structured in under a minute. Backend: – Designed a distributed platform based on a microservice architecture: a set of independent ser-vices on which the product is built: a unified authorization and access management service: roles and permissions scoped to user workspaces; a permissions and feature management ser-vice: enabling and restricting product features by roles and pricing plans; a notification delivery service; a file service: secure upload and validation of user files; billing: pricing plans, subscrip-tions, invoicing and payment acceptance in compliance with UAE tax requirements; data hubs: a unified repository of legislation and case law with versions, links between instruments, and ci-tations; smart search across legal instruments. Result: together with the team, brought the platform to production: access and permissions, pricing plans, subions, billing, files, notifications, search, a unified data hub, observability. Stack: Python 3.14 Asyncio, Litestar, Granian, PostgreSQL, SQLAlchemy, Alembic, OpenSearch, Redis, NATS JetStream, GCloud Storage, Temporalio, GKE, PyMuPDF, Scrapling, Playwright, Vertex AI, PaddleOCR, MinerU, OTEL.

Company profile: Development of Enterprise RPA solutions for end-to-end business process automa-tion: from ETL pipelines to predictive analytics. Designed and implemented the architecture of an internal product for automating order search, which became the company’s main sales channel. – Live monitoring: developed a system for collecting data from key freelance platforms in real time, minimizing the delay between an order being published and being detected. – Data normalization: implemented an NLP-based pipeline for extracting named entities and transforming unstructured order deions into a unified format. – Predictive analytics: implemented complex two-level scoring logic: AI-based assessment of task relevance to the company’s technology stack and experience; automated KYC analysis of clients to filter out non-target requests. – Business process automation: created a Telegram bot for managing presale. When a target or-der is detected, the system notifies the team and allows initiating an automatic response in one click. Result: reduced the response time to a new order from several hours to several seconds. Full process automation drove growth in deal conversion through instant contact with priority clients. – Developed and maintained more than 50 scrapers of varying complexity for external clients in E-commerce, FinTech, Real Estate, Social Media. – Unified the data collection architecture and introduced a library of reusable components, which sped up the development of new solutions by 30%. Stack: Python 3.12+ Asyncio, FastAPI, Uvicorn, Pydantic, SQLAlchemy, PostgreSQL, MySQL, Redis, Min-IO, RabbitMQ, Temporalio, playwright, HTTPX, Telethon, Docker-compose.

Company profile: Development of a high-load platform for collecting and processing sports analytics in real time for B2B clients. – Designed and implemented a system for aggregating data from dozens of competitor websites protected by sophisticated anti-bot systems. – Migration from Selenium to Playwright: initiated the move of the crawling stack to Playwright. Optimized server resource consumption and ensured more stable operation in headless mode, which made it easy to scale the number of replicas for collection. – Anti-bot strategies: implemented protection bypass techniques: browser fingerprint rotation, use of Stealth plugins, and dynamic proxy rotation. Brought the data collection success rate up to 97.5%. Result: built a fault-tolerant storage of historical and live data, reducing the lag behind market data by 60%. Participated in the development of a video analytics service that extracts advanced metrics from match video streams in real time. – Object tracking: integrated YOLO / OpenCV models for detecting and tracking key objects: the player, the ball, and court markings. – Spatial analytics: developed algorithms for calculating unique metrics: ball landing heat maps, movement speed, zone coverage, and shot statistics. – API and real-time delivery: designed the REST API architecture for clients and implemented a WebSocket server with a subion mechanism. Clients receive analytics with <200 ms laten-cy immediately after a frame is processed. – Task queues: introduced Celery + RabbitMQ for asynchronous processing of video events, ensur-ing stable handling of peak loads during fast-paced tournaments. – Database optimization: optimized heavy analytical queries in PostgreSQL and added a Redis caching layer. As a result, reduced the generation time of complex client reports from 10 sec-onds to <500 ms. Stack: Python 3.8+, FastAPI, Selenium, OpenCV, PyTorch, PostgreSQL, SQLAlchemy, Rab-bitMQ, Celery, Docker-compose, WebSockets.