Ramil Kaneevmchomak
en
Back home

Case proof base

Detailed cases: context, architecture, challenges and result

On the homepage, cases work as a showcase. Here each project is expanded as a proof base: why it was needed, what was built, where the technical risks were and what working result came after launch.

10 cases

Telegram bots, mini apps, neural networks, crypto automation, server services and websites in one format.

Estimate a similar project
01Telegram / Subscription automation

Subscription Bot - selling subscriptions with automatic access delivery

Telegram bot for selling subscriptions: plans, four payment methods, automatic access delivery and a control panel for operators.

Subscription bot Telegram flow and admin dashboard preview - preview_rec
01 / 09

Task context

The client needed to sell subscriptions right inside Telegram, without delivering access manually 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 intake, a control panel, reminders and handling of non-standard cases were required.

What was implemented

Step-by-step purchase: plan, location, payment method, invoice and access delivery.

Control panel for plans, users, payments, payment methods and broadcasts.

Automatic jobs: subscription expiration, reminders and bulk broadcasts.

Fast work thanks to cached settings that refresh after changes in the control panel.

Connections and work logic

Telegram Bot API

Rapira

CryptoBot

ParityPay

PostgreSQL

Workflow

UserTelegram botaiohttp backendPostgreSQLpayment webhooksadmin panel

Screenshots and interfaces

real or sample mockups
Telegram bot
main menuplanspayment provider selectionaccess delivered
Admin panel
payments by providerplan editorbroadcastspayment settings
Engineering
purchase FSMAPScheduler jobswebhook flow map

Challenges and solutions

Several payment methods

All payments are brought to one format so the operator sees statuses the same way, regardless of the payment method.

Automatic access without chaos

The purchase is split into clear steps, and access is delivered only after a confirmed payment.

Settings without a developer

Plans, locations and payment methods are changed right in the control panel, without reinstalling.

Result

Operators only handle disputed cases and support.

Plans, locations and payment methods are managed through the control panel.

The user receives access immediately after successful payment.

02Mini App / Telegram Mini App / exchanger

SapsanEx - a mini app for an exchanger inside Telegram

Mini app for an exchanger: rate calculation, request creation, status tracking and notifications - all inside Telegram.

SapsanEx Telegram Mini App calculator and admin workflow preview - preview_rec
01 / 05

Task context

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

On a phone, users were lost between the bot, browser and status page.

The third-party exchanger service was inconvenient to connect to and gave no clean request structure.

What was implemented

Mini app with a calculator, request creation and history.

Server gateway to the exchanger service with data validation.

Telegram bot for launching the app and status change notifications.

Automatic cancellation of stuck requests and status updates right in the app.

Connections and work logic

Telegram WebApp initData

Premium Exchanger API

Telegram Bot API

PostgreSQL

Workflow

UserTelegram WebAppReact Mini AppFastAPI gatewayPremium Exchanger APIPostgreSQLaiogram bot

Screenshots and interfaces

real or sample mockups
Mini App
calculatorrequest confirmationstatus pollingrequest history
Telegram notifications
request createdstatus changedtimeout auto-cancel
Engineering
HMAC initDatatyped API wrapperWebApp flow map

Challenges and solutions

Login without a separate account

Login happens automatically and securely through Telegram, so the user does not need a separate account.

Inconvenient external service

The connection to the exchanger service is hidden behind a clean layer, so the app works with understandable data.

Stuck requests

An automatic job closes expired requests, while the user sees the current status right in the app.

Result

Calculation, request creation and tracking remain inside Telegram.

Statuses arrive through bot notifications and update inside the app.

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

03AI / AI bot / e-commerce

Seedream Bot - clothing try-on with a neural network in Telegram

Telegram bot for virtual try-on: product upload, settings, the Seedream neural network, payment via Telegram Stars and YooKassa, plus a control panel.

Seedream try-on bot with before and after generation preview - preview_rec
01 / 10

Task context

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

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

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

What was implemented

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

Connection to the Seedream neural network with request validation and result delivery.

Telegram Stars and YooKassa payments with balance accounting.

Control panel for users, plans, payments and manual credits.

Connections and work logic

Seedream API

Telegram Stars

YooKassa

Telegram Bot API

PostgreSQL

Workflow

UserTelegram botFastAPI backendSeedream APIpaymentsPostgreSQLadmin panel

Screenshots and interfaces

real or sample mockups
Telegram bot
product uploadparametersAI resultgeneration history
Payments
Stars invoiceYooKassa checkoutgeneration balance
Admin panel
userspaymentstariffsmanual credits

Challenges and solutions

A neural network as a product, not a demo

Generation is embedded into a clear user flow with balance, history and plans.

Two payment methods in one bot

Telegram Stars are handy for quick small purchases, while YooKassa is for packages and subscriptions.

Operator control

The control panel shows payments, users and credits without database access.

Result

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

The store can manage plans and balances without developer involvement.

The solution is ready for product demos and reuse in other online stores.

04AI / AI Telegram bot

AI Reply Assistant - Telegram bot with 7 smart scenarios

Smart assistant inside Telegram: seven scenarios, the GPT-4o neural network, payment via YooKassa, friend invites and backup access channels for stable work.

AI reply assistant Telegram bot and moderation pipeline preview - preview_rec
01 / 05

Task context

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

Access to the neural network had to survive slow or failing channels.

Trial limits, packages, receipts and invite bonuses had to stay consistent.

What was implemented

Telegram scenario menu with context upload and answer variants.

Seven scenarios, each with its own request preparation and answer parsing.

Payment via YooKassa, a free trial and an invite program.

A set of backup access channels with health checks and disabling of slow ones.

Connections and work logic

OpenAI GPT-4o

YooKassa

Telegram Bot API

PostgreSQL

HTTP proxy pool

Workflow

UserTelegram botscenario routerprompt builderproxy poolGPT-4oPostgreSQL

Screenshots and interfaces

real or sample mockups
Telegram bot
main menuscenario selectioncontext uploadanswer variants
Payments
packagesYooKassa checkoutreferral screen
Engineering
prompt builderproxy rotationYooKassa webhook

Challenges and solutions

Neutral and clear interface

Scenarios are framed as a universal messaging assistant, without questionable wording.

Unstable external access

The system itself picks a working channel, temporarily disables failing ones and skips slow ones.

Clear payment

Trial, packages, receipt and invites are connected to user limits.

Result

The user stays in Telegram and gets answer variants without manual copying.

The bot itself picks a working access channel to the neural network.

Payments and bonuses go into a single user balance.

05Crypto / Crypto trading automation

ByBit Trading Bot - automated crypto trading 24/7

Trading automation: connection to the Bybit exchange, a strategy based on volume spikes and price acceleration, risk control and Telegram reports.

Bybit trading bot market scanner and alert dashboard preview - preview_rec
01 / 03

Task context

Manual watching was too slow for the number of trading pairs.

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

The operator needed Telegram reports and visibility into open trades.

What was implemented

Market scanner on the Bybit exchange with selection by liquidity.

Trading strategy based on volume spikes and price acceleration.

Risk control: stop signals, profit taking and a limit on the number of open trades.

Telegram notifications about trades, positions and daily profit or loss.

Connections and work logic

Bybit REST API

Bybit WebSocket

CoinPaprika API

Telegram Bot API

PostgreSQL

Workflow

Bybit market datascannerstrategyrisk managerorder layerPostgreSQLTelegram reports

Screenshots and interfaces

real or sample mockups
Strategy
volume spike chartprice accelerationpositions table
Telegram
entry signalposition closedaily report
Engineering
scanner logstrategy codeposition manager

Challenges and solutions

Over 300 trading pairs

The scanner is separated from the strategy so market selection does not mix with the entry rules.

Testing without risk

Test mode lets everything run on real data without real trades on the account.

Trade control

The system tracks the limit of open trades, stop signals and exit conditions.

Result

The bot does not need constant manual watching of the market.

All key events go to Telegram.

Test mode reduces risk before turning on real trading.

06Crypto / DEX trading / ML pipeline

EPS Bot - trading on the Solana blockchain with trainable models

Research system for exchanges on the Solana blockchain: collecting price data, training models, strategy, placing trades on the blockchain and Telegram reports.

Solana trading bot Telegram alerts and backend architecture preview - preview_rec
01 / 03

Task context

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

Models and settings had to be replaceable without redoing the connections to exchanges.

The technical details of the blockchain needed to be isolated from the strategy.

What was implemented

Connections to exchanges for collecting price data.

Storage for samples and history.

A set of trainable models of different types.

A layer for placing trades on the Solana blockchain and Telegram reports.

Connections and work logic

Raydium API

GeckoTerminal API

Solana RPC

Telegram Bot API

PostgreSQL

Workflow

DEX datacollectortraining pipelinestrategytx_toolsSolana RPCTelegram reports

Screenshots and interfaces

real or sample mockups
ML and market
OHLCV signalstraining curvesmodel compare
CLI
collect poolstrain modelbacktest report
Engineering
LSTM snippetstrategy thresholdRaydium tx

Challenges and solutions

Data first, then the model

Data collection and training are separated so the model can change without redoing the connections to exchanges.

Different models

The system allows comparing several models and settings for different market conditions.

Trades on the blockchain

Trade placement is isolated so the strategy does not depend on the technical details of the blockchain.

Result

The system became the basis for the next crypto projects and automations.

Models and settings can be changed without breaking the whole system.

Telegram reports give the full picture without a separate panel.

07Web / Crypto exchanger redesign

Frax - design update for a WordPress crypto exchanger

Visual update for a WordPress exchanger: new design, a convenient calculator on phones and careful work on top of the existing plugin.

Frax real estate website redesign before and after preview - preview_rec
01 / 05

Task context

The client already had a working exchanger on WordPress with a specialized plugin.

The plugin stored rates, directions, limits, requests and partner connections.

A full move was expensive and risky, so the look had to be updated on top of the existing logic.

What was implemented

New unified style and reworked page templates.

Main screen and exchange calculator, convenient on phones.

Neat embedding of the plugin data into the new design.

Lightweight styling code without complex tools, so the site is easier to maintain.

Connections and work logic

WordPress

Exchange plugin

PHP templates

shortcodes

CSS/JS

Workflow

UserWordPress pageshortcode wrapperexisting pluginexchange backendoperator

Screenshots and interfaces

real or sample mockups
Before / after
old heronew heromobile calculator
New design
exchange directionsrequest formFAQ / rules
Engineering
PHP shortcode wrapperCSS component layer

Challenges and solutions

Do not break the plugin

The exchange logic was not rewritten: the new design simply wrapped the existing plugin data.

Calculator on phones

Fields, buttons and action order were redesigned for comfortable one-handed use.

Client support

Styling and templates were kept simple, so the client can maintain the site without complex tools.

Result

The site became visually closer to the 2026 exchanger market.

On phones, placing a request became clearer and larger.

The existing exchange logic stayed stable and familiar to the client.

08Backend / Landing / lead automation

Tech Rise Academy - a site with leads straight to Telegram

Multi-page academy site: leads are checked and sent to the owner in Telegram within seconds.

TechRise Academy landing page and Telegram lead bot preview - preview_rec
01 / 05

Task context

The academy had no site and no separate budget for a client management system.

Leads were getting lost between direct messages, email and manual chats.

A lightweight solution was needed that could be maintained without a complex server part and database.

What was implemented

Main page and two course pages.

A single settings file for links, contacts and the offer.

A small lead intake service with data validation.

Sending a ready lead card to Telegram.

Connections and work logic

FastAPI

Pydantic

Telegram Bot API

Docker Compose

Nginx

Workflow

Visitorlanding formFastAPI endpointTelegram Bot APIowner chatNginxDocker

Screenshots and interfaces

real or sample mockups
Landing
heroprogramscourse pagelead form
Lead flow
Telegram lead cardconfig.jslead handler
Engineering
Docker ComposeNginxFastAPI endpoint

Challenges and solutions

Without a client system

A Telegram chat became the single place for all leads, without a separate database or paid systems.

Edits without a developer

Contacts, links and the offer were moved into a separate settings file.

Fast reaction

The lead goes to the owner immediately after the form is checked.

Result

The owner gets the lead right away in the familiar Telegram.

Text changes are made precisely in the settings file.

The site works without heavy technology and a separate client system.

09Backend / Niche web app

gym_progres - workout tracker with auto-save

Personal site for tracking workouts: auto-save, ready templates via a link and progress history.

Gym Progres training tracker and analytics dashboard preview - preview_rec
01 / 05

Task context

Notes and spreadsheets are a poor fit for regularly logging sets and progress.

Ready apps are overloaded with extra features and subscriptions.

A lightweight product was needed with ready templates via a link and saving without a button.

What was implemented

App pages that are built on the server.

Auto-save as you type, without a Save button.

An exercise catalog that can be expanded without a restart.

Ready workout templates and adding them without duplicates.

Connections and work logic

FastAPI

Jinja2

Alpine.js

PostgreSQL

Docker Compose

Workflow

UserSSR pageAlpine.jsJSON APIdatabasetemplatesprogress view

Screenshots and interfaces

real or sample mockups
Web app
dashboardworkout formexercise catalogprogress chart
Templates
public templateimport confirmshare modal
Engineering
auto-save snippettemplate importSSR + JSON architecture

Challenges and solutions

Auto-save without surprises

Changes save automatically as you type, so the user simply enters data.

Templates without duplicates

Adding a template is set up so existing exercises are not duplicated.

Lightweight app without heavy technology

A simple approach gave a fast interface without bulky tools.

Result

Sets are saved without a button and manual control.

Templates can be shared with a regular link.

History and charts give a quick overview of progress.

10AI / AI education product

SkillUp - a learning platform powered by a neural network

A complete learning service with a neural network: getting to know the user, a personal development plan, a progress tree and background task processing.

SkillUp AI learning tree and onboarding chat preview - preview_rec
01 / 02

Task context

Ordinary neural network answers turn poorly into a clear step-by-step plan.

Ready learning-plan sites are too general and do not account for the user's level.

A complete scenario was needed: getting to know the user, building a plan, progress, explanations and tests.

What was implemented

A site with an interactive progress tree.

A server part with a database and work logic.

Background processing of heavy neural network tasks with progress display.

The Claude neural network for getting to know the user, the plan, explanations and tests.

Connections and work logic

Anthropic API

Celery

Redis

PostgreSQL

Docker Compose

Nginx

Workflow

UseronboardingAI pipelineCelery workerClaude APIdatabaseprogress tree

Screenshots and interfaces

real or sample mockups
Product
onboardingplan buildertreequiz
AI pipeline
onboarding chatplan generationnode explanationquiz generation
Engineering
Celery taskcanvas layoutClaude API call

Challenges and solutions

A plan instead of a wall of text

The neural network's answer is turned into a plan of 4 levels with steps, statuses and order.

Long neural network tasks

Building the plan runs in the background so the site shows progress and does not freeze.

Convenient progress tree

The position of the steps is calculated automatically, without storing coordinates in the database.

Result

The user answers a couple of questions and gets a personal learning plan in a few minutes.

Progress is clearly visible through open and locked steps.

The neural network's work is split into building the plan, explanations and tests.