Feature/ai model training

[DongDuong2001]

This pull request introduces an end-to-end training and inference pipeline for a custom AI assistant model tailored for software development tasks and technical Q&A. It includes scripts for synthetic dataset generation, a configurable training setup, a FastAPI-based inference server, and all necessary dependencies. The most important changes are grouped below:

1. Dataset Generation and Training Pipeline

• Added generate_dataset.py and a more advanced generate_dataset_backup.py to synthesize ~4000 examples (task creation and tech Q&A) in TinyLlama chat format for model fine-tuning. These scripts use templates and randomization to create diverse, structured prompts and responses. [1] [2]
• Added training_config.yaml for reproducible training runs, specifying model, LoRA, dataset, and generation hyperparameters.

2. Inference Server

• Introduced inference/server.py, a FastAPI app for serving the fine-tuned model with endpoints for text generation and structured task creation. The server loads LoRA adapters if available, applies chat formatting, and exposes CORS for integration.

3. Documentation and Dependencies

• Added a detailed README.md with setup, training, inference, and hardware requirements to guide users through the pipeline.
• Created requirements.txt listing all dependencies for training, inference, and utilities, ensuring reproducibility.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Review	Updated (UTC)
lab68dev-platform-1ds5	Ready	Preview, Comment	Jan 22, 2026 4:09pm

[Copilot]

Pull request overview

This pull request introduces a complete AI model training and inference pipeline while removing the existing RAG (Retrieval-Augmented Generation) system. The PR replaces browser/server-based RAG embeddings with a standalone Python-based training pipeline for fine-tuning TinyLlama for software development tasks.

Changes:

• Removed RAG-based knowledge base system including embeddings service, indexing scripts, and related dependencies
• Added Python-based AI model training pipeline with LoRA fine-tuning for TinyLlama
• Introduced FastAPI inference server for serving the fine-tuned model

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 15 comments.

Show a summary per file

File	Description
scripts/index-knowledge.js	Removed knowledge base indexing script (RAG removal)
lib/services/rag-service.ts	Removed RAG embeddings and document search service
app/api/chat/route.ts	Removed RAG context retrieval from chat API
package.json	Removed RAG-related dependencies and indexing scripts; version bump to 0.1.1
ai-model/train.py	New training script with LoRA configuration and 4-bit quantization
ai-model/requirements.txt	Python dependencies for training and inference
ai-model/inference/server.py	FastAPI server for model inference with generation endpoints
ai-model/data/generate_dataset.py	Synthetic dataset generator for training examples
ai-model/data/generate_dataset_backup.py	Incomplete backup dataset generator (truncated)
ai-model/config/training_config.yaml	Centralized training hyperparameters and model configuration
ai-model/README.md	Setup and usage documentation for the AI training pipeline

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

@DongDuong2001 I've opened a new pull request, #18, to work on those changes. Once the pull request is ready, I'll request review from you.