AI-native developer · Thailand, UTC+7

I build AI assistants, bots and data tools. Scoped, tested, shipped.

You bring the input and the goal. I design the logic and the checks; AI agents write the code.

Send a briefSee the work ↓
lead-finder · runningillustrative data
The pattern behind a live project: chats → AI filter → a queue a human confirms.

AI assistants

Answers grounded in your real data · from $430

Telegram bots

Support, orders, team workflows · from $210

Data & scrapers

Messy source → clean data · from $500

Websites

Design → fast live site · from $280

Selected work

How each system works, drawn out. Diagrams use illustrative data.

AI assistant · Jewelry brand

A product finder that can’t invent products

It talks to shoppers like a consultant and recommends real pieces from the catalog — every suggestion is verified, so it never shows a product that doesn’t exist.

How it works
  1. Clarify first. The dialog scores how well it understands the visitor and asks more only when it’s unsure.
  2. Validate IDs. Suggested products are checked server-side; unknown IDs go back to the model as “not found”.
  3. Real cards. Name, price and image come from the catalog, never from the model’s text.
Next.jsClaude APIPythonTelegram
Live assistant ↗Website + AI assistant
Brand site + assistant · Skincare

A brand store with an AI assistant wired into the CRM

Catalog, journal and an admin the team runs without a developer. The assistant talks to customers and hands them to the team when needed.

How it works
  1. Shared session. Widget and full chat use one CRM session key on the device.
  2. Answers in parts. Long replies arrive block by block, so the visitor isn’t left waiting.
  3. Recovers. If the connection drops, the dialog resumes instead of starting over.
Next.jsPostgreSQLCRM APIPython bot
Live site ↗Store + assistant + admin
Data pipeline · Legal consultant

Turns busy group chats into a steady flow of leads

It reads the chats 24/7, spots people who actually need the service — not just talk about it — and hands them to the consultant as ready lead cards.

How it works
  1. Collect. Messages from chosen public Telegram groups.
  2. Classify. A topic pre-filter, then the model checks intent, category and urgency; output is schema-validated.
  3. Queue. A bot card with the original message, link and status buttons.
PythonTelethonaiogramClaude
Private system · demo on a call
E-commerce ops · Running for two brands

Marketplace reviews answered in seconds, not hours

Every new review and question lands in Telegram with a ready reply in the brand’s voice. Tweak it with a note, publish with one tap — you stay in control.

How it works
  1. Pull. Unanswered reviews and questions via the marketplace API.
  2. Draft. A reply following the brand’s rules; the operator can ask for a rewrite with a note.
  3. Publish. Sending is a separate human action, plus a summary of what’s still open.
PythonWildberries APIOpenAITelegram
Private system · demo on a call
Content pipeline · Events editor

Dozens of channels in, one clean event feed out

Announcements from many channels become structured event cards, duplicates merged. The editor just picks what to publish.

How it works
  1. Extract. Tell events from ordinary posts; pull title, date, place, link.
  2. Merge. Match by title, date and place, then by wording for same-day events.
  3. Moderate. A ranked queue with cards and approve buttons for the editor.
PythonhttpxaiogramSQLite
Private system · demo on a call

Pricing

Pick a base, add what you need. You get a fixed scope — what’s in, what’s out, how long.

1 · Base
2 · Add-ons
How we work
01BriefYour input, goal and 3 examples of a right result
02ScopeFixed milestone, price and acceptance checks
03BuildDemo early, tested against your examples
04HandoverCode, setup notes, deployment
Reply within a day

Tell me what goes in and what should come out.

rodionwork@yandex.ru Telegram ↗
10–15 h / weekFocused part-time capacity
UTC+7Overlap with Europe mornings & Asia
Fixed scope firstNo open-ended hourly surprises
You own the codeSource, notes and handover