Which model do you use?

The one that fits your task and budget, usually Claude or Gemini, or a provider you already use, through your own account. The eval set lets you compare models later without rebuilding.

Can you guarantee accuracy?

No. Language models make mistakes. You get a measured score on your own examples, and risky actions are gated so a person approves them.

Will my data be used to train models?

The work runs on your own API account, so your provider’s data-use terms apply. You choose the provider and plan, and the spec lists exactly what data the feature sends.

Do I need a call before I pay?

No. The scope, price and delivery time are on this page, and everything happens in writing. If you want to talk live, calls cost $100 per 15 minutes, paid upfront.

Who owns what you deliver?

You do. The code, videos and files delivered under this offer belong to you once paid.

Fixed-price offerAdd one AI feature or agent to your product, at a fixed price

You know where AI would save time in your product or your team’s day, but nobody has had the time to build it properly. One feature or agent is specified with you, built on the model API you choose, and tested against your own examples so you can see how well it works before anyone relies on it.

Pay $2,500 & start

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts.

Check fit by email

Simon does the work himself
5.0 from 127 Fiverr reviews, checked Oct 30, 2026
Paid to App369 LLC through Stripe. Receipt by email, invoice with your company name on request.

$2,500 fixed price, 10 business days after the one-page spec is agreed. You pay your model provider directly for usage.

Price checked October 25, 2026.

No call needed. Ask a question first by email at contact@app369.com or with the contact form; we reply by email, usually within one business day.

AI Feature or Agent Build

One LLM feature or agent wired into your app or internal tools, such as a reply drafter, a document extractor or an agent that calls your API, with an eval set so you can see how well it works before you ship it.

$2,500

fixed price, 10 business days

You get

A one-page spec you approve first: the task, inputs, outputs, the tools it may call and how success is measured
One feature or agent built on the Claude or Gemini API, or another model API you already use, on your own provider account
Tool access limited to the actions you approve, with write actions gated
An eval set of up to 50 cases from your examples, with a scored run before handover
Logging of prompts, tool calls and token cost
A pull request with a README and handover notes
One revision round

Not included

Training or fine-tuning models
A guaranteed accuracy score
A second feature or agent (quoted separately)
Model usage costs, which you pay your provider directly

Delivery

10 business days after the spec is agreed and repo access and test keys arrive.

Simon builds Workspace369’s AI features and MCP server, and has published open-source reference code for agent reliability testing.

Pay $2,500 & start

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts.

After you pay

Simon emails you within one business day to confirm and send the access checklist. You can also send it straight away to contact@app369.com. What the work needs:

The feature you want, in a few sentences
20 or more example inputs with the outputs you would call good
Repo access, plus test-environment API keys for the model provider and any tools it should call

You get a one-page spec to approve first. Delivery takes 10 business days after the spec is agreed.

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts.

Not sure it fits? See who it fits, or email a question about this offer; we reply by email, usually within one business day.

This fits when

Your team copies text between tools and rewrites it by hand
Support or sales replies follow patterns an assistant could draft
You need structured data pulled out of emails, PDFs or forms
You want an agent that can take actions through your own API, with limits
A demo works, but nobody has measured how often it gets things right

How it works

Agree the spec
A one-page spec: the task, the inputs and outputs, the tools it may call, and how success is measured. Nothing is built until you approve it.
Build the eval set
Up to 50 test cases from your real examples (with personal data removed), so quality is measured, not guessed.
Build and wire it in
The feature or agent goes into your codebase on your provider account, with write actions gated and every prompt, tool call and token cost logged.
Score and hand over
A scored eval run, a pull request, a README and notes on how to extend it.

What the failure handling looks like

From the Claude Agent Reliability Kit, a public demo repo by Simon: an agent that reads insurance claim emails and hands structured data to a workflow. Its demo runs four invented claims, and three of them go wrong.

Sample claim	What goes wrong	What the agent does	Result
Rear-end collision	Nothing	Looks up the policy, then submits the claim as a tool call checked against a fixed schema	Done, coverage confirmed
Hail damage to a roof	The API answers 429 (rate limited) on the first call	The queue waits for the retry-after time, then tries again	Done, coverage confirmed
Brake failure	The amount comes back as “about 1,800 dollars”, and the policy has lapsed	Sends the validation errors back once, and the model returns 1800	Done, coverage not confirmed
Physiotherapy costs	The policy number doesn’t exist	The lookup error goes back to the model as a tool error instead of crashing the run	Done, coverage not confirmed
The first claim sent again	A duplicate submission	Its idempotency key matches, so it is ignored	Not processed twice

Rear-end collision
What goes wrong
Nothing
What the agent does
Looks up the policy, then submits the claim as a tool call checked against a fixed schema
Result
Done, coverage confirmed
Hail damage to a roof
What goes wrong
The API answers 429 (rate limited) on the first call
What the agent does
The queue waits for the retry-after time, then tries again
Result
Done, coverage confirmed
Brake failure
What goes wrong
The amount comes back as “about 1,800 dollars”, and the policy has lapsed
What the agent does
Sends the validation errors back once, and the model returns 1800
Result
Done, coverage not confirmed
Physiotherapy costs
What goes wrong
The policy number doesn’t exist
What the agent does
The lookup error goes back to the model as a tool error instead of crashing the run
Result
Done, coverage not confirmed
The first claim sent again
What goes wrong
A duplicate submission
What the agent does
Its idempotency key matches, so it is ignored
Result
Not processed twice

Sample only: invented claims, a public demo, not a client’s data. The demo uses a scripted stand-in for the model, so it shows the plumbing, not accuracy. A real build adds a scored eval run on up to 50 of your own examples.

Read the code on GitHub

Built before

Work Simon has shipped that uses the same kind of build. Every project links to something you can check.

2026Open source by Simon Dziak

Claude Agent Reliability Kit

AI agents

A reference build for making a Claude tool-use agent dependable in production: loop caps, schema-enforced output, a durable retry queue, prompt caching and per-call tracing.

TypeScriptClaude API

GitHub

Questions before you pay

Which model do you use?: The one that fits your task and budget, usually Claude or Gemini, or a provider you already use, through your own account. The eval set lets you compare models later without rebuilding.
Can you guarantee accuracy?: No. Language models make mistakes. You get a measured score on your own examples, and risky actions are gated so a person approves them.
Will my data be used to train models?: The work runs on your own API account, so your provider’s data-use terms apply. You choose the provider and plan, and the spec lists exactly what data the feature sends.
Do I need a call before I pay?: No. The scope, price and delivery time are on this page, and everything happens in writing. If you want to talk live, calls cost $100 per 15 minutes, paid upfront.
Who owns what you deliver?: You do. The code, videos and files delivered under this offer belong to you once paid.

Fiverr reviews / 95 on app work

What buyers said about working with Simon

Read 126 of the 127 reviews

“Outstanding work! He went above and beyond to make an amazing app for me! I will definitely be working with Szymon again!”
Mark, United States, April 2021. App development on Fiverr
“I work with Szymkn to fix bugs and implement small features and he always delivers on time. thanks again”
A buyer in the United States, January 2021. App development on Fiverr
“This seller did an amazing job again. He always knows exactly how to complete any crazy ideas I bring to him. Thank you”
A buyer in the United States, December 2020. App development on Fiverr

Read the whole story

Sixteen five-star reviews from one buyer: mobile apps, app design and a Shopify store (Mark, United States)
A large custom app build, delivered on the back end and the front end (A buyer in the United States)

Ready to start?

Pay online and send what the work needs by replying to the Stripe receipt. Simon, who builds and runs the SaaS Workspace369, does the work himself.

Pay $2,500 & start

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts.

Related fixed-price offers

Fixed scope, fixed price, paid upfront. The work happens in writing; no call needed.

Looking for something else? See all fixed-price offers

Fixed-price offerAdd one AI feature or agent to your product, at a fixed price

AI Feature or Agent Build

You get

Not included

Delivery

After you pay

This fits when

How it works

Agree the spec

Build the eval set

Build and wire it in

Score and hand over

What the failure handling looks like

Built before

Claude Agent Reliability Kit

Questions before you pay

What buyers said about working with Simon

Read the whole story

Ready to start?

Further reading

Related fixed-price offers