Detailed cases: context, architecture, challenges and result

The client needed a Telegram sales funnel for digital subscriptions without manual credentials delivery in direct messages.

The manual process did not scale: operators mixed up plans, locations and payment statuses.

A simple chat bot was not enough: payment webhooks, an admin panel, reminders and controlled edge cases were required.

FSM purchase flow: plan, location, payment provider, invoice and access delivery.

Jinja2 admin panel for plans, users, payments, providers and broadcasts.

APScheduler for subscription expiration, reminders and background broadcast jobs.

Settings and button cache with invalidation after admin panel changes.

Telegram Bot API

Rapira

CryptoBot

ParityPay

PostgreSQL

Operators only handle disputed cases and support.

Plans, locations and providers are managed through the admin panel.

The user receives access immediately after successful payment.

Customers came from Telegram, but the key exchange steps happened on an external website.

The mobile funnel lost users between the bot, browser and status page.

The third-party exchanger API used form-data and lacked a convenient typed request model.

React + TypeScript Mini App with calculator, request creation and history.

FastAPI gateway over the exchanger API with Pydantic models.

aiogram bot for WebApp launch and status change notifications.

Background timeout cancellation and status polling inside the interface.

Telegram WebApp initData

Premium Exchanger API

Telegram Bot API

PostgreSQL

Calculation, request creation and tracking remain inside Telegram.

Statuses arrive through bot notifications and update inside the Mini App.

A local request journal stores history and reduces dependence on the external API.

The goal was to turn AI try-on into a paid Telegram product with a clean user flow.

Users needed history, balance and understandable limits, not a one-off generation command.

The operator needed visibility into users, payments, tariffs and manual credits.

Telegram bot flow for product upload, parameters, generation and result history.

Seedream API integration with request validation and result delivery.

Telegram Stars and YooKassa payments with balance accounting.

FastAPI admin panel for users, tariffs, payments and manual credits.

Seedream API

Telegram Stars

YooKassa

Telegram Bot API

PostgreSQL

A user can buy generations and receive AI try-on results inside Telegram.

The store can manage tariffs and balances without developer involvement.

The architecture is ready for product demos and further e-commerce integration.

The product needed several AI scenarios without turning the bot into a confusing prompt editor.

Access to the model had to survive slow or failing proxy nodes.

Trial limits, packages, receipts and referral bonuses had to stay consistent in the database.

Telegram scenario menu with context upload and AI answer variants.

Prompt builder for seven use cases with shared safety and formatting rules.

YooKassa checkout, referral screen and user limits.

Proxy pool with cooldown for failed nodes and rotation for slow ones.

OpenAI GPT-4o

YooKassa

Telegram Bot API

PostgreSQL

HTTP proxy pool

The bot can serve multiple AI reply scenarios from one clean menu.

Payments and limits are connected to user state.

Proxy rotation improves model access stability.

Manual monitoring was too slow for the number of spot instruments.

Signal logic needed to be testable without immediately placing real orders.

The operator needed Telegram reports and visibility into positions.

Bybit WebSocket/REST market scanner with filtering.

Volume Spike + Price Acceleration strategy.

Position manager with limits, stops and exit conditions.

Telegram signals, close notifications and daily reports.

Bybit REST API

Bybit WebSocket

CoinPaprika API

Telegram Bot API

PostgreSQL

The system scans markets and sends signals without manual watching.

Dry-run and logging make strategy checks safer before live mode.

Telegram reports keep the operator informed about entries, exits and health.

The goal was not only to trigger trades, but to test the full path from data to execution.

Models and thresholds had to be replaceable without rewriting API clients.

Solana RPC and DEX transaction details needed isolation from strategy code.

OHLCV collection for Solana pools.

Training pipeline with model comparison and backtest reports.

Strategy threshold layer for different market regimes.

Raydium / Solana transaction tools separated from strategy logic.

Raydium API

GeckoTerminal API

Solana RPC

Telegram Bot API

PostgreSQL

The R&D loop covers data, training, backtest and execution.

Models and thresholds can be changed independently.

On-chain execution is separated from strategy logic.

The business logic already worked inside WordPress and could not be replaced quickly.

The old interface reduced trust and was uncomfortable on mobile.

The client needed maintainable templates without heavy build infrastructure.

New landing and exchange direction screens.

Mobile calculator and application form improvements.

PHP shortcode wrapper around existing plugin output.

CSS component layer for maintainable visual updates.

WordPress

Exchange plugin

PHP templates

shortcodes

CSS/JS

The exchanger received a more trustworthy interface.

The rollout did not require backend replacement.

The client can maintain the visual layer in WordPress.

The academy needed to launch pages and start collecting leads quickly.

A separate CRM would add overhead for the first version.

Contacts, links and offer details had to be editable without code changes.

Multi-page landing with hero, program sections, course page and lead form.

FastAPI endpoint for validation and Telegram delivery.

config.js for contacts, links and offer settings.

Docker Compose and Nginx deployment setup.

FastAPI

Pydantic

Telegram Bot API

Docker Compose

Nginx

The academy can collect leads immediately after launch.

The owner gets structured Telegram notifications.

Basic content settings can be changed without touching backend logic.

A training tracker needed a fast interface without heavy client-side infrastructure.

Set data had to persist while the user was still editing.

Shared templates needed import without duplicates.

FastAPI SSR dashboard, workout form and exercise catalog.

Alpine.js auto-save with debounced JSON API updates.

Public templates, import confirmation and share modal.

Progress history and charts.

FastAPI

Jinja2

Alpine.js

PostgreSQL

Docker Compose

Workout entry is fast and resilient to missed save clicks.

Templates can be shared and imported safely.

The product stays lightweight and easy to maintain.

The product had to turn a vague learning goal into a concrete path.

Long AI tasks needed progress feedback instead of a frozen interface.

The roadmap had to remain structured for explanations, statuses and quizzes.

Onboarding chat that collects goals and current level.

AI roadmap generation with tiered nodes.

Interactive progress tree with node statuses.

Celery/Redis background tasks and Claude API calls.

Anthropic API

Celery

Redis

PostgreSQL

Docker Compose

Nginx

The learner receives a structured roadmap instead of generic advice.

Long AI generation does not block the interface.

The tree can grow with new node types and quizzes.