Mchomak

Hamster

A workout-tracking web app for sets, weights, exercises and progress. Every input autosaves — no “Save” button. Problem. Notes and spreadsheets become messy, while fitness apps add ads, subscriptions and unnecessary features. PDF plans are tedious to transfer manually. Solution. FastAPI + Jinja2 power an SSR interface; Alpine.js sends debounced API updates as the user types. A 40+ exercise catalog is stored in data/exercises.json, cached and extendable without container restarts. Public links let students import templates without duplicates. Bodyweight exercises use profile weight automatically. Sessions run through SessionMiddleware, config through pydantic-settings, deployment through Docker Compose. Result. Choosing a template, logging sets and viewing progress fits on one screen - without ads or subscriptions. Templates can be shared without registration. Stack: Python 3.12, FastAPI, SQLAlchemy, PostgreSQL, Jinja2, Alpine.js, Tailwind CDN, Docker Compose. https://gymprogress.ru

A frontend redesign of the frax.cc crypto exchanger without replacing WordPress or rewriting the exchange plugin. The goal was to modernize the visual layer, mobile UX and calculator while preserving the existing business logic. The plugin already handled rates, directions, limits, orders and partner APIs, so migration would have been expensive and risky. Instead, WordPress templates were rebuilt, a new visual system was introduced, grids, typography and the core exchange flow were cleaned up. Plugin output was embedded into the new markup through hooks and shortcodes. The redesign is live in production: the site feels current, the mobile flow is usable, and the backend stayed stable and familiar for the client. frax.cc

A Telegram bot on aiogram 3 that prepares reply options from a text message or a conversation screenshot. The user stays inside Telegram: chooses a scenario, sends context and receives several GPT-4o-powered variants. The project combines seven AI scenarios, separate prompt-builder and response-parser modules, tone and role personalization, YooKassa payments, a referral program and proxied OpenAI access. Model calls go through an HTTP proxy pool with healthchecks, cooldown after failures and slow-node rotation, so one bad proxy does not block the whole flow. The MVP with payments is ready for production rollout: the bot selects a working proxy automatically, clears payments with fiscal receipts and returns replies in one Telegram window. @chat_help_12bot

A Telegram bot for selling digital subscriptions with automated access delivery after payment. The public framing stays neutral: subscription product, payments, admin panel and operator workflow automation. One aiohttp service handles the bot webhook, payment-system webhooks and a Jinja2 admin panel. Purchase is an FSM flow: plan, location, payment provider, invoice and access delivery through integrations. APScheduler tracks expirations, sends renewal reminders and runs broadcasts. In the admin panel, operators manage plans, payments, users, providers and broadcast scenarios. The result is an automated funnel from plan selection to access delivery, where a human is only needed for edge cases and support. @sapsanshopbot

A Telegram Mini App for the sapsanex.cc exchanger: users calculate a rate, create an order and track status without leaving Telegram. The app consists of a React+TS frontend, FastAPI backend, aiogram bot and PostgreSQL in Docker Compose. Authentication uses Telegram WebApp initData with HMAC-SHA256, with no separate login. The backend wraps the third-party Premium Exchanger API into a typed layer and also keeps its own order journal. The bot sends status-change notifications, while the Mini App handles polling and timeouts. The result is a complete exchange flow inside Telegram: calculation, order creation, payment wait, auto-cancel for stale orders and final status notification. @sapsan_robot https://sapsanex.cc/

A multi-page landing page for an IT academy with two course pages and automated lead delivery to the owner's Telegram. The project solves the intake flow without a CRM, database or heavy framework. The frontend is plain HTML/CSS/JS: home page, course pages, forms and a shared config.js where the client can edit links, contacts and offer URLs without a developer. The FastAPI backend accepts /api/lead, validates data with Pydantic and sends a lead card to Telegram through the Bot API. Deployment is Docker Compose + Nginx. The owner receives each application seconds after form submission and can react inside the familiar Telegram chat, without checking email or paying for a separate CRM. https://techrise.ru/index

A Telegram bot for AI virtual try-on: a shop uploads a product photo, and Seedream 4.0 generates an image of a model wearing that item without a separate photoshoot. The aiogram 3 bot guides the user through an FSM flow: product upload, generation parameters, payment and result delivery. Monetization has two lanes: Telegram Stars for quick in-Telegram purchases and YooKassa for packs/subscriptions. A FastAPI admin panel on Jinja2 + Bootstrap shows users, balances, payments, plans, manual top-ups and exports. The result is a finished e-commerce product: bot, generation, two payment systems and an operator admin panel. For new img2img cases, only the model provider has to change.

SkillUp is a web service where AI builds a personal learning plan for any topic and shows progress as a living tree instead of a checklist. The frontend uses Next.js 14, TypeScript, Zustand, TanStack Query and Framer Motion; the backend uses FastAPI, PostgreSQL, Celery and Redis. The AI layer runs on the Anthropic API: one model assembles the 4-tier plan, another handles onboarding, explains nodes and generates quizzes. Tree nodes have `locked / active / done` states, while coordinates are computed on the frontend from tier and order_index, so layout can evolve without migrations. In a few minutes the user goes from onboarding to an interactive roadmap where progress is visual: finish a block, and the branch blooms.

A client-built trading bot for the Bybit exchange. It monitors 300+ spot pairs around the clock, detects momentum moves using a Volume Spike + Price Acceleration strategy and opens and closes positions autonomously. Three-layer architecture: a WebSocket/REST scanner with CoinPaprika liquidity filtering, a strategy module that computes volume anomalies and price acceleration, and a risk manager with stop controls and a position-count limit. Telegram notifications cover every trade and a daily P&L summary. Dry-run mode for validating the strategy on live data before going live.

An autonomous trading bot for Raydium DEX on Solana. It collects pools and OHLCV data, trains ML models on historical candles and executes trades from strategy signals without manual clicks. The pipeline has four layers: market collection through Raydium API v3 and GeckoTerminal, PostgreSQL storage, a PyTorch/scikit-learn ML module with LSTM, GRU, CNN, Transformer and Random Forest models, then strategy logic plus tx_tools for Solana transactions. Telegram alerts report trades, failures and daily PnL. The project proves the full loop from data to model to trade to report, and became a technical base for later crypto and automation cases.