AI Backend Overhaul: Model Integration & Optimization
Budget: $30 – $250 USD
Project Brief: AI Model Switching and Backend Optimization
Background:
Our AI-powered application uses an Apollo/GraphQL backend and multiple integrated AI models for reasoning, judgment, transcription, and OCR.
The previous developer managed model setups manually. I now need the system restructured so that I (the client) can switch models directly in the backend, once properly trained.
Objectives
Backend Model Switching:
Implement a modular setup so models can be switched or added easily by editing backend configuration (e.g., .env or MySQL model table).
No hard-coded parameters — the system should pull model settings dynamically (model name, endpoint, temperature, token limits, etc.).
After setup, provide me a short walkthrough or documentation on how to:
Switch model providers (OpenAI, Gemini, Granite, etc.)
Add new ones in the future
Restart or reload the backend to apply changes
File Audit and Refactor:
Review and document the following files that define or call AI models:
src/Helpers/AskAI.ts
src/Helpers/AITools.ts
src/cron/AIJudgement.ts
src/cron/AITasks.ts
Ensure model calls are handled via a single central function or model loader, to make switching consistent across the app.
GraphQL Backend Integration:
Add a mutation for restarting failed judgment processes.
Integrate it into the admin-back-office:
Create a secure endpoint for restart
Add mutation under the graphql folder
Modify components/documents.tsx to use it for the “Restart” button
Include input sanitization and permissions.
Recommended Models to Test/Integrate:
Google Gemini 2.5 Pro – general reasoning, strong context handling
IBM Granite 4.0 – fast, lightweight, cost-effective, low resource use
Qwen3-VL 4B – for vision/OCR acceleration, smaller and faster
DeepSeek OCR – advanced lightweight vision-language model
DeepSeek OCR Paper
Performance & Security Guidelines:
Use indexed fields for any database status checks.
Keep API keys and endpoints in .env or a secured config file.
Sanitize all GraphQL inputs and ensure permissions for admin-only actions.
Comment your code well to help future edits.
Deliverables
Modular backend model configuration system
Working GraphQL restart mutation
Documentation or short training session (to show me how to switch models manually)
Optional: a summary table or JSON config listing all available models and their parameters
Background:
Our AI-powered application uses an Apollo/GraphQL backend and multiple integrated AI models for reasoning, judgment, transcription, and OCR.
The previous developer managed model setups manually. I now need the system restructured so that I (the client) can switch models directly in the backend, once properly trained.
Objectives
Backend Model Switching:
Implement a modular setup so models can be switched or added easily by editing backend configuration (e.g., .env or MySQL model table).
No hard-coded parameters — the system should pull model settings dynamically (model name, endpoint, temperature, token limits, etc.).
After setup, provide me a short walkthrough or documentation on how to:
Switch model providers (OpenAI, Gemini, Granite, etc.)
Add new ones in the future
Restart or reload the backend to apply changes
File Audit and Refactor:
Review and document the following files that define or call AI models:
src/Helpers/AskAI.ts
src/Helpers/AITools.ts
src/cron/AIJudgement.ts
src/cron/AITasks.ts
Ensure model calls are handled via a single central function or model loader, to make switching consistent across the app.
GraphQL Backend Integration:
Add a mutation for restarting failed judgment processes.
Integrate it into the admin-back-office:
Create a secure endpoint for restart
Add mutation under the graphql folder
Modify components/documents.tsx to use it for the “Restart” button
Include input sanitization and permissions.
Recommended Models to Test/Integrate:
Google Gemini 2.5 Pro – general reasoning, strong context handling
IBM Granite 4.0 – fast, lightweight, cost-effective, low resource use
Qwen3-VL 4B – for vision/OCR acceleration, smaller and faster
DeepSeek OCR – advanced lightweight vision-language model
DeepSeek OCR Paper
Performance & Security Guidelines:
Use indexed fields for any database status checks.
Keep API keys and endpoints in .env or a secured config file.
Sanitize all GraphQL inputs and ensure permissions for admin-only actions.
Comment your code well to help future edits.
Deliverables
Modular backend model configuration system
Working GraphQL restart mutation
Documentation or short training session (to show me how to switch models manually)
Optional: a summary table or JSON config listing all available models and their parameters