Ravish

Key Skills

Software

Other

Top Subject Matter

Retrieval-Augmented Generation (RAG) for document question answering

LLM fine-tuning with LoRA & PEFT

Secure AI / prompt injection detection and red teaming

Top Data Types

Text

Top Task Types

Question Answering

Fine-tuning

Red Teaming

Freelancer Overview

Retrieval-Augmented Generation (RAG) Question Answering System. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include N, A, and Other. Education includes Bachelor of Technology, M B D College (2023) and Bachelor of Technology, Parul University (2027). AI-training focus includes data types such as Text, Computer Code, and Programming and labeling workflows including Question Answering, Fine-tuning, and Red Teaming.

Labeling Experience

PromptArmor – AI Firewall for Prompt Injection Detection

OtherRed Teaming

Developed an AI security system to detect and block prompt injection attacks in LLM applications. Built real-time threat-detection pipelines to identify jailbreak attempts, malicious prompts, and hidden instruction attacks. Implemented rule-based and NLP-based filtering with a risk-scoring mechanism to allow, sanitize, or block suspicious prompts before they reach the model. • Designed prompt risk analysis and classification using rule-based and NLP approaches • Implemented controls to sanitize or block high-risk prompts in secure LLM workflows • Evaluated and refined detection logic for jailbreak and hidden-instruction patterns • Built the service using a transformer-based NLP approach integrated into an API

2026 - Present

LLM Fine-Tuning using LoRA & PEFT

OtherTextFine-tuning

Fine-tuned pre-trained Large Language Models on custom datasets using LoRA and PEFT techniques. Reduced computational requirements while improving task-specific performance. Performed tokenizer and prompt-formatting work to support effective transformer training and evaluation. • Trained and validated the model while tracking training/validation loss metrics • Implemented LoRA/PEFT fine-tuning workflows in PyTorch and Hugging Face • Optimized tokenizer settings and prompt formatting for consistent input structure • Conducted evaluation using loss metrics to assess model improvements

2025 - 2025

Retrieval-Augmented Generation (RAG) Question Answering System

TextQuestion Answering

Built a Retrieval-Augmented Generation (RAG) question-answering pipeline for document-based QA tasks. Integrated vector-embedding-based document retrieval to generate contextual responses. Optimized embedding and retrieval workflows to improve response relevance and quality. • Implemented REST APIs using FastAPI for serving QA outputs • Created an interactive UI using Streamlit for question/answer exploration • Worked with vector embeddings and retrieval logic to ground answers in documents • Evaluated and refined retrieval/embedding steps based on response relevance

2025 - 2025

Education

Work History