You will get an AI feature shipped into your existing app, with cost controls and evals
Project details
Adding AI to a product is easy to demo and hard to ship. The demo works on three examples; the shipped version has to handle the other nine hundred, stay inside a budget, and not embarrass you when a user asks something odd.
I build the shipped version.
WHAT YOU GET
• The feature integrated into your existing codebase, not a separate demo app
• Cost controls: token budgets, caps, and what this costs per 1,000 users
• Graceful failure: what the user sees when the model is slow, down, or wrong
• Documentation of the prompts and why they are written that way
WHAT I BUILD MOST OFTEN
• Question answering over your own documents, tickets or knowledge base
• Summarisation and extraction from messy text
• Classification and routing for support tickets, leads, or moderation
• Agent workflows that call your existing APIs and take real actions
WHY EVALS MATTER
An AI feature degrades silently: a model update makes it worse and nothing throws an error. Advanced includes a fixed set of inputs with known-good outputs, so you can prove it still works before you ship a change.
If a plain database query solves your problem better, I will say so instead of selling you a model call.
I build the shipped version.
WHAT YOU GET
• The feature integrated into your existing codebase, not a separate demo app
• Cost controls: token budgets, caps, and what this costs per 1,000 users
• Graceful failure: what the user sees when the model is slow, down, or wrong
• Documentation of the prompts and why they are written that way
WHAT I BUILD MOST OFTEN
• Question answering over your own documents, tickets or knowledge base
• Summarisation and extraction from messy text
• Classification and routing for support tickets, leads, or moderation
• Agent workflows that call your existing APIs and take real actions
WHY EVALS MATTER
An AI feature degrades silently: a model update makes it worse and nothing throws an error. Advanced includes a fixed set of inputs with known-good outputs, so you can prove it still works before you ship a change.
If a plain database query solves your problem better, I will say so instead of selling you a model call.
AI Algorithms
Large Language Model, Transformer ModelAI Applications
AI Chatbot, AI-Enhanced Classification, Conversational AI, Natural Language Generation, Natural Language UnderstandingAI Development Language
PythonAI Models
GPT-4What's included
| Service Tiers |
Starter
$350
|
Standard
$850
|
Advanced
$1,800
|
|---|---|---|---|
| Delivery Time | 5 days | 10 days | 14 days |
Number of Revisions | 1 | 2 | 3 |
AI Model Integration | |||
Batch Normalization | - | - | - |
Database Integration | - | ||
Detailed Code Comments | |||
Image Upscaling | - | - | - |
MLOps | - | - | - |
Model Deployment | - | - | - |
Model Documentation | |||
Model Monitoring | - | ||
Model Testing & Optimization | - | - | |
Model Tuning | - | - | - |
Natural Language Processing | |||
NLP Tokenization | - | - | - |
Pre-Training | - | - | - |
Prompt Engineering | |||
Setup File | - | - | - |
Source Code |
Frequently asked questions
About Haitao
AI Integration & Automation Engineer | Next.js, Python, OpenAI API
Beijing, China - 2:12 am local time
Most of my work falls into three areas:
AI INTEGRATION
Adding real AI capability to existing products: OpenAI API, LangChain, RAG pipelines over your own data, agent workflows, and the evaluation setup that keeps them from silently degrading.
WEB DEVELOPMENT
Production applications in Next.js, React, TypeScript and Node.js. Architecture through deployment, including the parts most people skip: structured data, performance, and search visibility.
AUTOMATION & DATA
Python that does the tedious thing reliably: web scraping, migrating data between systems with mismatched schemas, ETL pipelines, and third-party API integrations.
What I actually optimize for is turnaround. I run a solo studio with a heavily AI-assisted workflow, so work that would traditionally take two weeks often ships in three to five days. The time saved goes into testing and edge cases, not into billing more hours.
How I work:
- Fixed price wherever possible. You know the cost before I start.
- I will tell you if your problem does not need me. Some things really are a 20-minute fix with a tool that already exists.
- Timezone UTC+8, with overlap available for European mornings and US evenings.
Send me the specifics of what you are trying to build and I will tell you honestly whether it is a good fit and what it would take.
Steps for completing your project
After purchasing the project, send requirements so Haitao can start the project.
Delivery time starts when Haitao receives requirements from you.
Haitao works on your project following the steps below.
Revisions may occur after the delivery date.
Scope and cost estimate
I read your sample data and tell you what is realistically achievable, what it will cost per 1,000 users, and whether a model call is the right tool at all - before any work starts.
Prototype on your real data
I build the first working version against your own examples, not toy inputs, and show you where it succeeds and where it fails.