I build production AI integrations.

Claude/OpenAI agents, Telegram bots, scraping pipelines, automation. From prototype to production in days, not months. Four systems currently running in production.

About

I'm Walter Balialin — software engineer based in Ukraine. I connect large language models to real business systems: not demos, not slideware — production code that handles real users, real data, and real failure modes.

My stack centers on Python (FastAPI, aiogram, Telethon, asyncio), the Claude Agent SDK and OpenAI APIs, PostgreSQL, and the unsexy infrastructure work that makes these systems survive contact with real users — caching, rate-limit handling, observability, deploy.

I work directly with founders and small engineering teams. No middlemen, no process theater. If you can describe the integration in one paragraph, I can usually estimate it in two.

Python FastAPI aiogram Telethon Claude Agent SDK OpenAI API MCP PostgreSQL SQLite Docker n8n Hetzner Cloudflare

Selected Work

Fintech · LLM Web Search · Python

Live equity screener with LLM catalyst search

700-ticker fundamental screening engine with cached yfinance microservice and LLM-powered catalyst analysis on the filtered subset. Daily automated run for a private investment client.

Python yfinance LLM Web Search PostgreSQL Telegram Bot Docker
Screening: minutes → seconds. Robust against Yahoo Finance rate-limits via explicit cooldown + negative-cache layers. LLM stage is the only remaining bottleneck — exactly where it should be.
Mobile · Mini App · LLM analysis

App Store review intelligence — Telegram Mini App

Submit an App Store ID. Bot scrapes reviews country-by-country, runs LLM sentiment and thematic analysis, returns a structured Word report plus an Excel raw-data export.

Python FastAPI aiogram Mini App OpenAI SQLite Docker
Hours of manual review reading → 2-minute end-to-end pipeline. Self-serve workflow; no analyst-in-the-loop.
Request live demo →
SMB Automation · Low-code · Operations

Payment registration automation

Telegram bot finite-state machine replacing manual data entry into a legacy accounting system. State persists in n8n DataTables — no external database required for this footprint.

n8n Telegram Bot API DataTables Webhooks Inline keyboards
~20 hours/week of accountant time recovered. At fully-loaded labor cost, that's roughly $2,000/month in operational capacity returned to higher-value work. Payback in weeks.

More Projects

Bot Ostap — Telegram market & content intelligence platform. Telethon userbot for passive multi-channel listening, n8n-based spam classifier, query bot with /today, /market, /ask commands over the cleaned corpus.

Get in touch

Working on something that needs an AI integration done properly? Reach out — I respond within 24 hours during business days.