Aug 2024 – Present · 1 yr 7 mos · Bengaluru, Karnataka, India

• Built AI systems to fully automate customer review and image moderation, replacing manual workflows with real-time LLM pipelines.
• Text Review Moderation:
• ● Using Gemma2 LLM, I created classifiers that filtered irrelevant, offensive, or low-quality reviews, ensuring only clear, product-relevant English content was published.
• ● Achieved 96.2% accuracy, vs. 60% manually.
• ● Reduced moderation time from 20 days to real-time, expanded capacity 20X, and cut costs by 83%.
• ● Saved 2,400+ hours monthly, freeing teams for higher-value work.
• ● Built models for relevance, inappropriate content, language filtering, and keyword enrichment, boosting high-quality reviews by 40%.
• Image Moderation:
• ● In the next phase, I implemented Gemma3 12B LLM pipelines to automate image validation across 12 product categories.
• ● Delivered 90–97.5% accuracy across Groceries, Electronics, Home, Fashion, Beauty, and more.
• ● Developed systems to detect identity info (PII), inappropriate visuals (e.g., nudity, violence), screenshots, and packaging photos.
• ● Created adaptive prompts and rejection logic for category-specific validation.
• Impact:
• ● Combining text and image automation laid the foundation for a fully automated review pipeline, dramatically improving product credibility, customer trust, and compliance.
• ● This transformation turned a slow, manual process into a scalable, real-time workflow that ensures high-quality content with minimal effort and cost.
• Tech Stack:
• Python, Jupyter, Gemma2 9B & Gemma3 12B LLM, Ollama, LangChain, Prompt Engineering, Pandas, SQL, Excel, Tableau

May 2024 – Aug 2024 · 3 mos · Bengaluru, Karnataka, India

• ● During my internship at Reliance, I have gained hands-on experience in generative AI, ChatGPT prompt engineering, and LangChain basics.
• ● I successfully developed an Automated Question Answering System using web scraping, Retrieval-Augmented Generation (RAG), and the Llama 3 model with Ollama integration.