Contact information

You can get this candidate's contact information from https://www.work.ua/resumes/15441019/

Work experience

AI engineer

from 03.2026 to 07.2026 (5 months)
Vibe Development, Дистанційно (IT)

Sole engineer for the company's internal AI infrastructure. Designed the architecture, wrote the code, provisioned the servers, and operated the system in production from scratch with no pre-existing AI stack or team.

System architecture & production operations:
• Designed a multi-agent ecosystem of 7+ Python services sharing a common data layer (SQLite/MySQL + Google Sheets sync) and a unified Claude API layer, covering the full commercial funnel: market intelligence → lead discovery → qualification → personalized outreach → content publishing.
• Implemented cost-aware model routing across the whole system: roughly half of all Claude calls were served by Haiku (classification, lead scoring, relevance filtering, batch enrichment), with Sonnet reserved for generation-quality tasks, keeping inference cost proportional to the difficulty of each step rather than to volume.
• Deployed and operated all services on a Mac Mini production host via macOS LaunchAgents with auto-restart and full power-cycle recovery, monitored through launchctl; the system ran unattended for the duration of the engagement with no manual intervention required.
• Built error handling, retry, and multi-pass backfill logic so upstream failures (dead RSS feeds, scraper rate limits, API errors) degrade gracefully instead of silently producing empty output.

Lead generation & sales automation:
• Built a real-time Telegram pipeline (Telethon) extracting high-intent buyer leads from Bali real estate channels, with Claude-based intent classification and lead scoring that ranks prospects by purchase readiness and budget signals, separating qualified buyers from general chat noise.
• Automated context-aware first-touch outreach: Claude Sonnet generates personalized messages proposing relevant villa listings, with hot-lead alerts pushed to the sales team in Telegram. Surfaced up to 10 qualified buyer leads per week that had previously been found only by manually reading through chats.
• Built LinkedIn / Facebook / Twitter intelligence agents performing daily auto-discovery of real estate companies across Australia, Singapore, and Indonesia using rotating keyword strategies, plus an is_international classifier (Claude Haiku, batched) to filter the pipeline. Tech: Apify, PhantomBuster, parallel multi-actor scraping.
• Deployed a full-stack email outreach platform (Docker/Colima + PostgreSQL + Redis + FastAPI + React/Vite) on the production host; extended the codebase with a custom SMTP provider module to support the corporate mailbox, which the original Gmail-OAuth-only system couldn't handle.
• Built a B2B lead aggregation pipeline that scraped, deduplicated, verified, and enriched 1,200+ contacts into clean segmented lists across target APAC markets.

Content automation:
• Built an autonomous Telegram bot generating daily LinkedIn content from a 3-source pipeline (9 RSS feeds, sitemap scraping, Gemini keyword search) with Claude Haiku relevance scoring and backfill logic.

AI PROJECTS & HACKATHONS

from 09.2024 to now (2 years)
AI PROJECTS & HACKATHONS, Дистанційно (IT)

AI & SYSTEM PROJECTS

AI Video Repurposing SaaS — Multimodal Content Engine
Python, FastAPI, Qdrant, SQLAlchemy 2 (async), ARQ + Redis, PostgreSQL, React 19, Vite
• Architected an end-to-end multimodal pipeline transforming long-form video into vertical shorts via conversational agents.
• Built a 6-stage asynchronous indexing engine (whisper transcripts, visual embeddings, scene detection) using Qdrant vector database and Redis task queues for semantic clip retrieval.
• Designed schema-first architecture enforcing strict Pydantic models across 89 API routes, with dual-worker queue isolation to prevent latency spikes under high inference loads.

AI Legal Document Constructor (Diia Hackathon)
Python, FastAPI, Azure OpenAI, Groq, Llama 3.1, React, Pydantic
• Developed a conversational agent automating legal contract generation via dynamic multi-turn slot filling and intent routing.
• Built an AST-based .docx template ingestion pipeline using Llama 3.1 (Groq) for sub-second placeholder extraction and schema inference, backed by hybrid Pydantic/algorithmic validation guards.

SLM Translator — Custom 84M Transformer from Scratch
Python, PyTorch, NumPy, BPE, Seq2Seq Architecture
• Implemented an 84M-parameter Encoder-Decoder Transformer from scratch in PyTorch for EN-UA translation, trained on an 8.7M+ parallel sentence corpus.
• Built custom multi-head self-attention, sinusoidal positional encodings, a custom BPE tokenizer, learning-rate warmup, and gradient clipping, validating deep architectural understanding of modern LLMs.

High-Load Anomaly & Payment Prediction (INT20H 2026, 1st Place)
Python, LightGBM, Pandas, Feature Engineering
• Developed a production-grade classifier over 6.6M+ transaction records in a 22-hour hackathon, securing 1st place out of 53 teams.
• Implemented aggressive memory downcasting to operate within strict RAM limits, time-aware validation splits to prevent data leakage, and engineered cascading retry-logic features (F1 0.7985 val / 0.7584 test).


HACKATHONS & COMPETITIONS

• INT20H 2026 (BEST Kyiv, Mar 2026) — 1st place among 53 teams; engineered real-time anomaly detection and payment outcome prediction models under 22-hour deadline.
• INT20H 2025 (BEST Kyiv, Feb 2025) — Finalist; financial fraud detection, feature synthesis, and time-series forecasting.
• Diia.AI Contest (MinDigital x EPAM, Aug 2025) · Brainstack Data Camp (Aug 2025) · ML Week 2026 (Jan 2026).

Education

Київський політехнічний інститут імені Ігоря Сікорського

Прикладна математика
Unfinished higher, from 2023 to 2027 (4 years)

КПІ ім. Ігоря Сікорського

Факультет програмних систем та прикладної математики (ФПСПМ), Спеціальність 113 «Прикладна математика», Київ
Unfinished higher, from 2023 to 2027 (4 years)

Additional education and certificates

Всі курси на Coursera

починаючи з 2023
Certificate

Knowledge and skills

• Python
• LangChain
• ChromaDB
• Azure OpenAI
• Mathematics
• Statistics
• Algorithms
• OpenAI
• SQL
• PostgreSQL
• MySQL
• BigQuery
• SQLite
• FastAPI
• Flask
• REST API
• Docker
• Git
• GitHub
• React
• JavaScript
• Team colloboration
• Problem solving
• Self motivation
• Time management
• Continuous Learning
• Claude API
• Codex
• Antigravity
• Pydantic
• TypeScript
• Vite.js
• Tailwind CSS
• Redis
• Prompt Engineering
• Structured Outputs
• Web Scraping
• Machine learning
• Deep Learning
• RAG
• Gemini

Language proficiencies

• Ukrainian — fluent
• English — average

Additional information

- Graduated from high school with a gold medal for outstanding academic performance.
- INT20H 2026 (BEST Kyiv, March 2026) -- 1st place among 53 teams; anomaly detection and payment success prediction in 22 hours.
- INT20H (2025), BEST Kyiv Hackathon -- in the qualification stage, my team ranked 8th out of 44 teams, securing a place in the finals (Top 10). In the final round, we achieved 6th place overall. Developed machine learning models for fraud detection and time series forecasting.

===== Links =====
LinkedIn: [open contact info](look above in the "contact info" section)
GitHub: https://github.com/LatkoArtem
Kaggle: https://www.kaggle.com/latkoartem
Resume: https://drive.google.com/file/d/1E1BJP-St-x5BWS3dBbisByR8eREF94dF/view?usp=sharing