Search this site

Search this language. ↑ ↓ to choose, Enter to open, Esc to close.

    Services

    Small, well-defined engagements. No open-ended retainers.

    LLM feature sprint

    You have an app and an idea for an AI feature, but nobody on the team has shipped one. Two to three weeks later it is in your codebase, evaluated, and documented.

    From
    $1,200
    Timeline
    2–3 weeks from kickoff
    • One working feature integrated into your existing app
    • Prompt and retrieval code in your repository, reviewable, not locked in a no-code tool
    • A 20-question evaluation set built from your real cases, plus baseline and final scores
    • Deployment instructions and a 30-minute handover call, recorded
    • 30 days of bug fixes after delivery
    • Model and API costs (billed to your own account, so you keep the rate limits)
    • Redesigning surrounding UI beyond the component the feature needs
    • Ongoing monitoring, on-call, or feature expansion

    Data pipeline build or rescue

    The scraping vendor dropped your target sites, or the scraper someone wrote is silently returning empty rows. Fixed scope, alerting included, yours to keep.

    From
    $600
    Timeline
    1–2 weeks from kickoff
    • Collector for the agreed sites, with retry and backoff per site
    • Normalised output in your preferred format (Postgres, CSV, or a Google Sheet)
    • Scheduled runs on your infrastructure, or a documented cron setup
    • Failure alerts that fire on repeated empty results, not just exceptions
    • A runbook covering what to do when a target site changes its markup
    • Sites behind a login or CAPTCHA you do not have permission to access
    • Bypassing rate limits or terms of service
    • Long-term hosting — I set it up on your machine or VPS

    One automation script

    There is a manual task someone on your team does every week. It is small, it is boring, and it is exactly the wrong use of their time.

    From
    $400
    Timeline
    3–5 business days
    • One script that performs the described task end to end
    • A dry-run mode that shows what it would do without doing it
    • Logging you can read when something goes wrong
    • A README covering how to run it and what each failure message means
    • One revision round after you have actually run it on real data
    • Tasks requiring credentials or permissions you cannot grant
    • Ongoing maintenance after the revision round
    • Building a UI around it

    Questions

    What clients usually ask before starting.

    How do you charge?

    Fixed price per project, agreed before we start. You know the total up front. I do not bill hourly — hourly billing punishes you for my learning curve and punishes me for your scope changes.

    What does a project usually cost?

    Automation scripts start around $400. Data pipelines start around $600. LLM features start around $1,200. Anything with real integration risk gets a written estimate after a short call. Things that raise the price: multiple systems to integrate, unclear success criteria, or a deadline inside two weeks.

    How long does it take?

    The clock starts when I have access to what I need, not when we sign. Scripts are 3–5 business days, pipelines 1–2 weeks, LLM features 2–3 weeks. I take one project at a time, so the dates I quote are dates I can hold.

    How do we communicate, and in what timezone?

    Async by default — a shared channel and a short written update every working day. I am in UTC+8, which overlaps with Europe in my afternoon and with US mornings if we schedule ahead. Written English is fine; I do calls on request.

    Are you a team? Do you outsource?

    One person, no outsourcing. What you get is what I wrote. If a project needs skills I do not have, I will say so before you pay rather than quietly sub-contracting it.

    What do you not take on?

    Anything that needs me to bypass a site’s terms of service, CAPTCHAs, or rate limits. Crypto and gambling projects. Ongoing on-call support. Full product builds end to end — I am a good fit for one well-defined problem, not a whole platform.

    My project is very small, or much bigger than these packages

    Small is fine — a one-day script is a real project. For something bigger, the usual answer is to cut the first slice and ship that: one feature, one pipeline, one workflow. If that goes well, we scope the next slice. I do not take large fixed-price engagements up front.

    What if I am not happy with the result?

    Every package includes a revision round after you have run it on real data. If the first week’s prototype is clearly not going to work, we stop there and you pay only for that week. Beyond that: half up front, half on delivery, and the final half is not due until the acceptance criteria we wrote down are met.

    Who owns the code, and who maintains it afterwards?

    You own everything — code, prompts, documentation, accounts. It lives in your repository from day one, not in mine. 30 days of bug fixes are included. After that, maintenance is a separate small engagement if you want it; there is no lock-in if you would rather not.

    How do I pay you?

    For overseas clients: Wise, or the platform escrow if we met through Upwork or Fiverr. For China-based clients: 对公转账 or 支付宝. I can sign an NDA before the first call, and I invoice properly either way.