You will get a custom Claude AI agent, spec-first, in Python
Rising Talent
Rising Talent
Project details
Most AI agents look great in a 30-second demo and fall over on the second real input. I build the ones that stay up.
You work directly with me, not a rotating agency team. I write the agent and the infrastructure around it end to end, so the person who designs it is the person who handles your edge cases.
Spec-first means we agree on what the agent does, what goes in, what comes out, and how it behaves when something is wrong, all in writing, before any code exists. You read it, you push back, you sign off. No surprises at delivery.
Where this fits:
• A multi-step process a person runs by hand today (intake, review, routing, follow-up)
• Work that needs real decisions, not just if-this-then-that
• Anything where a wrong answer is worse than no answer, so the agent has to know when to stop and ask
Recent build: a 7-skill, 18-state Claude Code offboarding engine: 4 sub-agents, guardrail hooks, and an event log as the single source of truth so every run is auditable.
I'm Claude-native, so I know how to make it hold up in production. If a no-code tool got you most of the way but keeps falling apart whenever things get complicated or messy, that's when I take over.
You work directly with me, not a rotating agency team. I write the agent and the infrastructure around it end to end, so the person who designs it is the person who handles your edge cases.
Spec-first means we agree on what the agent does, what goes in, what comes out, and how it behaves when something is wrong, all in writing, before any code exists. You read it, you push back, you sign off. No surprises at delivery.
Where this fits:
• A multi-step process a person runs by hand today (intake, review, routing, follow-up)
• Work that needs real decisions, not just if-this-then-that
• Anything where a wrong answer is worse than no answer, so the agent has to know when to stop and ask
Recent build: a 7-skill, 18-state Claude Code offboarding engine: 4 sub-agents, guardrail hooks, and an event log as the single source of truth so every run is auditable.
I'm Claude-native, so I know how to make it hold up in production. If a no-code tool got you most of the way but keeps falling apart whenever things get complicated or messy, that's when I take over.
AI Algorithms
Large Language Model, Multimodal Large Language Model, Transformer ModelAI Applications
Conversational AI, Natural Language Generation, Natural Language Understanding, Text RecognitionAI Development Language
PythonAI Tools
Hugging FaceAI Models
OpenAI Codex, WhisperWhat's included
| Service Tiers |
Starter
$1,500
|
Standard
$3,000
|
Advanced
$5,000
|
|---|---|---|---|
| Delivery Time | 14 days | 21 days | 35 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 |
Optional add-ons
You can add these on the next page.
Fast Delivery
+$400 - $500
Additional integration / data source
(+ 3 Days)
+$400Frequently asked questions
1 review
(1)
(0)
(0)
(0)
(0)
This project doesn't have any reviews.
AM
Alan M.
Jun 15, 2026
Senior Data Engineer: Scale a Multi-Marketplace Catalog & Normalization Engine (PostgreSQL / Python)
Muhannad is great to work with and is a clear communicator which I value a lot. Will hire him again.
About Muhannad
Claude Code & AI Agent Engineer | Python Automation, LLM Pipelines
Cairo, Egypt - 4:25 am local time
Recent builds:
- A 7-skill, 18-state Claude Code offboarding engine: 4 sub-agents, hook-enforced guardrails, and a JSONL event log as the single source of truth so every run is auditable.
- Commission reconciliation across 7 insurance providers, each with its own format, delivered as reusable Claude skills the client now runs himself.
- A cross-marketplace product matcher at 99.2% precision, with an "abstain rather than mismatch" rule so it stops instead of guessing wrong.
- A 17-gate Python QA pipeline plus a two-pass LLM-judge review fleet for a fine-tuning dataset.
- A GEO upgrade for a news publisher, built with Claude Code and automated schema verification.
- A multi-source data pipeline delivered solo through an agentic Claude Code workflow: 3 databases → one audited, queryable product, solo.
When a no-code tool hits its ceiling (branching logic, state, real error handling), that's where I take over. I rebuild the workflow in code so it does what you actually meant, keeping the parts that worked.
I work best on projects where:
- The input and output are clear, even if the data in between is messy
- There is real API integration, data wrangling, or an AI/LLM workflow involved
- You need something built, tested and handed over, not just advised on
Why Claude-native matters: it's the model family I design around full time, so I know its failure modes and build around them instead of being surprised by them. If your project needs a different model for one step, I will tell you and wire it in. The goal is a system that holds up, not loyalty to a logo.
Stack: Claude Opus/Fable, GPT-5.5, Python, Node.js, OpenRouter, Tesseract, faster-whisper, Playwright, Pandas, SQL.
Steps for completing your project
After purchasing the project, send requirements so Muhannad can start the project.
Delivery time starts when Muhannad receives requirements from you.
Muhannad works on your project following the steps below.
Revisions may occur after the delivery date.
Spec and sign-off
I turn your goal into a written spec: the agent's job, its inputs and outputs, and how it behaves on bad data. You approve it before any code is written.
Build in Python
I build the agent and its infrastructure in your repo, on your keys, with real branching, state, and an abstain-rather-than-guess rule on the risky steps.