You will get a custom AI web application with FastAPI and LLM integration
Project details
You will get a custom AI web application designed around a real business workflow—not a generic chatbot template. I build production-focused systems using Python, FastAPI, LLM APIs, RAG, vector databases, structured outputs, tool calling, and third-party integrations.
My experience includes developing DEHAYZ, a full-stack AI SaaS platform; DEHA-CLI, a Codex-style multi-model coding agent; and multiple LoRA/QLoRA models and datasets published on Hugging Face. I can handle the complete process, including requirements analysis, system architecture, backend development, AI workflow design, testing, deployment, and documentation.
Depending on the selected tier and agreed scope, your solution can include document processing, internal knowledge assistants, workflow automation, AI agents, API integrations, voice or vision features, and database integration. You will receive clean source code, validation, error handling, setup instructions, and a system designed around your use case, budget, and infrastructure.
My experience includes developing DEHAYZ, a full-stack AI SaaS platform; DEHA-CLI, a Codex-style multi-model coding agent; and multiple LoRA/QLoRA models and datasets published on Hugging Face. I can handle the complete process, including requirements analysis, system architecture, backend development, AI workflow design, testing, deployment, and documentation.
Depending on the selected tier and agreed scope, your solution can include document processing, internal knowledge assistants, workflow automation, AI agents, API integrations, voice or vision features, and database integration. You will receive clean source code, validation, error handling, setup instructions, and a system designed around your use case, budget, and infrastructure.
Programming Languages
JavaScript, Python, TypeScriptCoding Expertise
Performance Optimization, SecurityWhat's included
| Service Tiers |
Starter
$199
|
Standard
$599
|
Advanced
$1,499
|
|---|---|---|---|
| Delivery Time | 7 days | 14 days | 21 days |
Number of Revisions | 1 | 2 | 3 |
Number of Pages | 1 | 3 | 3 |
Design Customization | - | ||
Content Upload | |||
Responsive Design | |||
Source Code |
Optional add-ons
You can add these on the next page.
Fast Delivery
+$79 - $349
Additional Revision
+$75
Additional Page
(+ 2 Days)
+$120Frequently asked questions
About Bahadir
Fullstack developer
Antalya, Turkey - 12:42 am local time
I do more than connect applications to an AI API. I fine-tune language models, build autonomous coding agents, create training datasets, design AI backends and turn prototypes into production-ready products.
I am the developer of DEHAYZ, a Turkish-first AI SaaS platform built as a complete AI workspace. It combines multi-model chat, my Turkish-focused ASLI model, PDF and document analysis, real PowerPoint generation, sourced web and academic research, voice and vision features, isolated workspace memory and automated response-quality checks.
I handled the product architecture, AI integrations, backend APIs, authentication, subscription logic, database layer, Linux/Nginx deployment and GPU infrastructure. This experience allows me to build not only an AI feature or experimental notebook, but an entire working product around the model.
I also built DEHA-CLI, an open-source Codex-style terminal coding agent written in TypeScript and Node.js.
DEHA-CLI supports Claude, OpenAI, DeepSeek, xAI, OpenRouter, Ollama and custom OpenAI-compatible endpoints. Its Plan → Code → Judge pipeline can use different models for planning, implementation and independent code review.
The agent can:
• Read, create and edit project files
• Execute shell commands and Python code
• Analyze existing codebases
• Browse websites with Playwright
• Capture screenshots and inspect interfaces with vision models
• Run API smoke tests
• Connect to MCP servers
• Maintain persistent conversation and project memory
• Automatically compress long contexts
• Block potentially destructive system commands
• Use local, cloud-hosted or self-hosted language models
This project demonstrates my experience in agentic workflows, tool calling, context engineering, multi-model orchestration, terminal application development, browser automation, AI-assisted code generation and production-focused safety controls.
My Hugging Face portfolio includes Turkish LoRA and QLoRA projects based on Qwen 9B, 4B and 2B models, Mistral/Ministral architectures and larger reasoning and mathematical models with up to 32 billion parameters.
I have hands-on experience with:
• Turkish language adaptation
• LoRA and QLoRA fine-tuning
• Supervised Fine-Tuning
• Instruction-following and conversational AI
• Reasoning and mathematical model training
• Turkish dataset creation and cleaning
• ChatML and multi-turn conversation formatting
• Tokenization and context-length management
• Checkpoint evaluation and failure-case correction
• LoRA adapter development and merging
• 4-bit and 8-bit quantization
• Memory-efficient GPU training
• Inference optimization
• Hugging Face model and dataset publishing
My work covers the complete LLM development lifecycle:
1. Understanding the target business use case
2. Choosing the appropriate base model and model size
3. Collecting, cleaning and structuring the training data
4. Designing prompts and conversation templates
5. Running LoRA, QLoRA or supervised fine-tuning
6. Evaluating checkpoints and identifying weak outputs
7. Improving the dataset based on evaluation results
8. Merging and quantizing the final model
9. Building an inference API or agent around the model
10. Deploying the complete system to production
What I can deliver:
• Custom LLM fine-tuning with LoRA, QLoRA or SFT
• Turkish or domain-specific language models
• Dataset collection, cleaning, formatting and quality control
• AI coding agents and autonomous multi-agent workflows
• Codex-style CLI tools and developer assistants
• RAG systems with document ingestion and persistent memory
• FastAPI inference APIs and AI backend integrations
• Model merging, quantization and inference optimization
• Web, PDF, image, voice and structured-data AI applications
• Local GPU, cloud GPU and Linux server deployment
• Debugging for CUDA, tokenizer, context-length and training issues
• Complete AI SaaS architecture and product development
Core technologies:
Python, PyTorch, Hugging Face Transformers, PEFT, TRL, Unsloth, BitsAndBytes, FastAPI, SQL, Redis, ChromaDB, TypeScript, Node.js, Playwright, Model Context Protocol, REST APIs, Nginx, Linux and GPU infrastructure.
I approach AI projects as complete engineering systems rather than isolated demos. Model selection, dataset quality, prompt format, context management, evaluation, cost, latency, security and deployment must be designed together from the beginning.
You will receive reproducible code, documented training parameters, clear communication and a solution designed around your actual business and deployment requirements.
Whether you need a fine-tuned model, an autonomous AI agent, a RAG assistant, a coding CLI or a complete AI SaaS product, I can take the project from data preparation and architecture to production deployment.
Steps for completing your project
After purchasing the project, send requirements so Bahadir can start the project.
Delivery time starts when Bahadir receives requirements from you.
Bahadir works on your project following the steps below.
Revisions may occur after the delivery date.
Requirements and Scope Review
I review your use case, files, integrations, risks, and success criteria. I then confirm the final scope, deliverables, timeline, and any technical dependencies.
Architecture and Working Prototype
I design the system architecture and build the first working workflow so the core AI behavior, output quality, latency, and integration feasibility can be validated early.