⚡ SYSTEM STATUS: ONLINE [LOCAL_CORE_RUNNING] 🤖 Model Layer: Qwen-0.5B (SFT Optimized) 🔋 Hardware Trace: Dynamic CPU/GPU Auto-Scale 🔒 Guardrails: Hybrid Routing Deterministic Parser 📊 Dataset: Custom .JSONL Stream Matrix 🧞 Portfolio-Genie-AI An End-to-End Local LLM Supervised Fine-Tuning (SFT) Pipeline with Hybrid Guardrails & Premium Streamlit UI 🪝 The Hook Why build this? Generative AI models are powerful, but they hallucinate when asked about highly specific, localized personal domains (like your exact resume details). Portfolio-Genie-AI solves this by implementing a full-cycle MLOps pipeline to fine-tune an open-weights LLM on custom datasets, wrapped inside a production-grade Hybrid Guardrail Architecture to completely eliminate hallucinations and deliver premium micro-interactions. 🚀 Key Features Local SFT Pipeline: Built using Hugging Face transformers and trl library with custom dataset tokenization schemas. Hybrid Guardrail Engine: Implements a deterministic response routing parser to completely mitigate model hallucination on critical factual data (skills, projects, education). Premium Web UI: A high-contrast, cyberpunk-themed dark tech chat interface featuring smooth CSS floating animations, custom hover-glow metrics, and seamless transition states. Hardware Optimized: Dynamic configuration logic that gracefully scales execution across CPU environments or standalone NVIDIA dedicated VRAM architectures. 🛠️ Tech Stack & Tooling AI/ML Frameworks: Hugging Face Ecosystem (Transformers, TRL, SFTTrainer, SFTConfig) Base Open-Weights Model: Qwen-0.5B / Custom Tokenizer Architecture Backend & MLOps: Python, Dataset Engineering (.jsonl), Git-Tracking Management Frontend Ecosystem: Streamlit (Custom Embedded HTML/CSS Transitions) 📂 Project Architecture Portfolio-Genie-AI/ ├── data/ │ └── portfolio_data.jsonl # Tokenized Ingestion Dataset ├── src/ │ ├── train.py # Main SFT Training Script │ └── test_genie.py # Guardrail & Logic Core Router ├── venv/ # Isolated Local Environment (Git-Ignored) ├── app.py # Premium Streamlit UI Script └── README.md # Project Documentation Matrix ⚙️ Quick Installation & Setup # 1. Clone the Architecture & Navigate inside git clone [https://github.com/ADITYA990-IND/Portfolio-Genie-AI.git] cd Portfolio-Genie-AI # 2. Configure Environment & Dependencies python -m venv venv # 3. Direct Pip Execution to install frameworks .\venv\Scripts\pip.exe install -r requirements.txt .\venv\Scripts\pip.exe install streamlit # 4. Trigger Local Fine-Tuning Execution .\venv\Scripts\python.exe src/train.py # 5. Run the Premium Chat Interface .\venv\Scripts\streamlit.exe run app.py 🛠️ Model Lifecycle & Ownership Developed & Trained By: Aditya Raj Chourasiya 🧞 Role: Computer Science & Artificial Intelligence Engineer Core Pipeline Focus: Supervised Fine-Tuning (SFT), Hybrid Deterministic Parsers, and Premium Dark Tech UI Architecture. Made with ⚡ by ADITYA990-IND — Core Local LLM Pipeline Framework © 2026