Hi, I'm noppruts (นพรุจ) from Thailand. I build AI agents, workflow automation and data tools: LINE and Telegram bots, n8n pipelines, satellite-radar analytics and quantitative research tooling.
Four projects that show the range. Details come from the projects' own documentation.
Turns open Sentinel-1 SAR imagery into per-plot flood and crop-damage estimates for Thai rice insurers and lenders. A rice-transplanting mask avoids flagging normal wet paddy as damage, and every result carries scene IDs, software versions, parameters and a confidence score so it can be defended in a claims dispute. Flood/damage module implemented; crop and carbon modules are interfaces for later.
n8n decides what to make and when (scheduling, topic selection, approval routing, analytics); a Python video factory turns a topic into a subtitled vertical video (script, video generation, TTS, FFmpeg) behind a local HTTP API with a restart-safe state machine. Nothing publishes without explicit approval.
One LINE Official Account serving two audiences: staff send fault codes or symptoms for an AI diagnosis and log completed jobs; customers link their LINE to their phone number and receive a reminder when the ~6-month service is due. A daily scheduler inside the same process pushes the reminders.
Full-stack system: FastAPI backend and Next.js dashboard where several AI agents analyse symbols, store predictions and compare them with later prices to track accuracy per timeframe. Builds daily reports and talks to users through a Telegram bot by command or natural language.
Filter by area.
Autonomous local CLI agent for test engineers: an agent loop that calls tools, plus ingestion, RAG, scheduler and a dashboard; packaged as an executable.
Plan → split into sub-tasks → call tools → verify → repeat until done, with a multi-agent orchestrator and a sandboxed workspace.
ReAct loop: capture the screen, ask a vision-language model for the next action, run it through GUI/shell tools, observe, repeat until the goal is met.
Runs three times a day on Bangkok time and routes each item through a Telegram bot for human approval before it goes out.
Photo of a trade slip → Thai+English OCR → Flex confirmation card → tax calculation → logged to an Excel workbook that doubles as the database.
Two importable n8n workflows for small online shops: a daily sales summary and a new-order alert with validation. Run end-to-end on a local n8n instance with test data.
Watches SET/mai stocks, generates signals from price and indicator thresholds, simulates trades in a paper portfolio, alerts via Telegram/LINE, Next.js dashboard. Live execution deliberately not implemented.
Market data and news sentiment (FinBERT/VADER) combined with EMA, RSI, ATR and Kelly sizing to output signals with dynamic stop-loss and take-profit levels.
Research workflow built around a written edge standard: net-of-cost expectancy, out-of-sample tests, sample-size minimums and regime analysis. Findings are not published.
An orange-cat character and supporting cast turned into LINE sticker, emoji and theme packs, short animated films and wallpapers. Films come from a deterministic frame-accurate HTML engine rendered with Playwright and FFmpeg, with synthesised audio normalised to -14 LUFS.
A product JSON and photo become a vertical clip with Thai voice-over in about a minute, without rendering a browser page per frame.
Mobile web tool that shows where each baht of a Shopee, TikTok Shop or Lazada sale goes, break-even price and break-even ROAS. All data stays in the browser. Free and paid editions are deployed on Cloudflare Pages.
Excel workbook for small shops with SUMIFS/INDEX-MATCH dashboards, low-stock alerts and charts. Every dashboard value was checked against an independent Python calculation.
Things you can open right now (Thai interface, built for online sellers).
See where each baht of a sale goes, the break-even price and break-even ROAS. Data stays in your browser.
Fee table for the three platforms with a worked example and an FAQ on checking unexpected deductions.
Three habits that show up across the projects above.
Numbers get checked against an independent calculation before they ship. Every fee table here cites its source and date.
Publishing, payments and anything irreversible wait for an explicit approval step. Automation drafts; people decide.
Start with the smallest version a real user can try, put it online, and let usage decide what gets built next.
Taken from the projects above, not a wish list.
Services listed on my Fastwork profile. Tell me what is slow, manual or unreliable in your process.
Short vertical videos (Reels, TikTok, Shorts) with Thai voice-over and subtitles.
LINE Official Account bots with automation: replies, reminders, staff tools.
Connect Sheets, LINE, Telegram and APIs so repetitive work runs itself.
Counting and quality-control from images, e.g. product counting.