👋 Welcome to My Portfolio

I’m Ravikiran Krishnaprasad, a Principal Engineer specializing in GenAI platforms, distributed systems, and cloud-native architecture on AWS. I design and build enterprise-scale AI systems using RAG, Agentic AI, and LLM orchestration, with a strong focus on reliability, evaluation, and scalability.

🚀 Impact & Achievements

Architected enterprise GenAI platform for SAP Joule using RAG and LLM orchestration
Generated $3.5M revenue through Payment Release Manager platform
Improved system throughput by 3× using event-driven architecture
Co-inventor of ML-based fraud detection system (US Patent filed)

🧠 Key Systems & Architectures

🏢 Enterprise GenAI Platform – SAP Joule Integration

Led architecture and enterprise adoption of GenAI platform using RAG, vector search, and LLM orchestration
Designed multi-agent workflows with governance guardrails for enterprise AI systems
Built scalable backend systems and evaluation pipelines for production AI deployment
Focus: reliability, hallucination mitigation, and enterprise-scale adoption

🧠 Multi-Agent Hallucination Detection & Correction System

Research Preprint + M.S. Thesis – LJMU

🔗 View GitHub Repository
📄 Read Research Square Preprint
🆔 ORCID

Description:

Designed and implemented a lightweight retrieval-grounded multi-agent framework for hallucination detection and correction in Large Language Model outputs
Developed the work as part of my independent MSc Artificial Intelligence and Machine Learning dissertation at Liverpool John Moores University
Refined the dissertation into a Research Square preprint, currently under journal review
Achieved ~0.93 F1 score across MedHallu and TruthfulQA datasets
Implemented retrieval grounding, threshold-based verification, and evidence-guided correction to improve factual reliability

Key Highlights:

End-to-end pipeline: Retrieval → Verification → Correction
Threshold-based hallucination verification using answer-to-evidence and question-to-evidence similarity
Lightweight alternative to expensive multi-LLM verification systems
Focus: hallucination mitigation, factual reliability, RAG evaluation, and responsible AI

📚 Publications

📝 Metrics-Driven Development in Cloud Enterprises

Summary: This paper presents a Metrics-Driven Development (MDD) framework for enterprise cloud environments, emphasizing real-time metrics to enhance performance, quality, and delivery.

🔗 Read the Tech Paper

🧠 Patents & Innovations

📄 ML-Based Fraud Detection System (Patent Filed – SAP Labs)

Co-inventor of ML-based fraud detection system for enterprise financial workflows
Enabled real-time transaction risk detection and compliance

Patent Details:

Application No: 18/903,764
Patent ID: 84058451
Patent Ref: 240333US01
Filed by SAP Labs India (US, Oct 2024)

Note: This is proprietary IP filed by SAP Labs. Included here as part of the inventor’s public contribution portfolio.

🚀 Applied GenAI & Machine Learning Systems

📚 ReAct-Powered RAG Assistant – Wikipedia Q&A Agent

Repo: react-rag-assistant

Models Used: OpenAI GPT-4o / GPT-3.5 (reasoning), OpenAI Embeddings (text-embedding-3-small), LlamaIndex ReActAgent Description: An AI agent that uses the ReAct (Reason + Act) pattern to dynamically index Wikipedia topics on user request and provide fact-based, context-grounded answers. Built with LlamaIndex, Chainlit, and OpenAI, the project demonstrates how ReAct reduces hallucinations, improves transparency, and enables complex multi-step tool use.

Use Case: One of the great ways to reduce hallucinations in LLMs by combining reasoning with retrieval — ensuring accurate, explainable, and grounded responses.

💬 Sentiment-Based Product Recommendation System – Ebuss

Repo: Sentiment-Based Product Recommendation System
Models Used: Logistic Regression, Random Forest, XGBoost, Naive Bayes, User/Item-Based Collaborative Filtering
Description: A hybrid recommendation engine that combines sentiment analysis of product reviews with collaborative filtering to deliver personalized top product suggestions. Built using NLP and ML techniques, and deployed via Flask on Heroku.

📄 SemanticSpotterGenAI – Gen AI - RAG-Based Generative Search for Insurance

Repo: SemanticSpotterGenAI
Models Used: RAG (Retrieval-Augmented Generation), OpenAI GPT, LangChain, LlamaIndex
Description: A RAG-powered AI system designed to simplify the search of complex insurance documents. It retrieves the most relevant content and generates intelligent, natural-language responses to user queries.

🧾 Mr.HelpMate AI – Gen AI - Search for Life Insurance Policies

Repo: Mr.HelpMate AI
Models Used: Sentence Transformers, FAISS, ChromaDB, OpenAI GPT (via Transformers)
Description: A Retrieval-Augmented Generation (RAG) based AI system that helps users understand life insurance documents by extracting key content and generating coherent, policy-specific answers to natural language queries.

🛍️ ShopAssist 2.0 – Gen AI - Enhanced Conversational AI with Function Calling

Repo: ShopAssist 2.0
Models Used: OpenAI GPT-4 (Function Calling API), Rule-based extractors
Description: An upgraded AI-powered chatbot that integrates OpenAI’s Function Calling to provide dynamic, structured laptop recommendations through a simplified, extensible architecture and refined conversational flow.

🧾 Automatic Ticket Classification – NLP-Based Complaint Categorization

Repo: Automatic Ticket Classification
Models Used: Non-Negative Matrix Factorization (NMF), Logistic Regression, Decision Tree, Random Forest
Description: An NLP-based system that automates the classification of customer complaints for a financial services firm using topic modeling and supervised ML. Enables faster ticket resolution by mapping complaints to relevant departments.

🩺 Disease-Treatment Mapping – Custom NER with CRF

Repo: Disease-Treatment Mapping Using Custom NER
Models Used: Conditional Random Fields (CRF)
Description: A custom Named Entity Recognition (NER) system for extracting disease-treatment pairs from unstructured medical text using CRF. Designed to support healthcare platforms in processing consultation notes and treatment history.

✋ Gesture Recognition – Deep Learning for Smart TV Controls

Repo: Gesture Recognition Project
Models Used: Conv3D, ConvLSTM, CNN + RNN Hybrid
Description: A video-based gesture recognition system trained to classify hand gestures for smart TV control using deep learning. Explores Conv3D and RNN architectures to accurately interpret user intent from spatiotemporal data.

🧬 Skin Cancer Detection – Melanoma Classification with Custom CNN

Repo: Skin Cancer Detection using Custom CNN
Models Used: Custom Convolutional Neural Network (CNN)
Description: A deep learning model built using TensorFlow to classify dermoscopic skin lesion images into various categories, including melanoma. Aims to assist dermatologists with early, automated diagnosis through high-accuracy image classification.

📞 Telecom Churn Prediction – Kaggle Competition (C-61)

Repo: Telecom Churn Prediction
Models Used: XGBoost, LightGBM, Random Forest, Logistic Regression
Description: A machine learning solution to predict churn probability in the telecom industry using behavioral and usage patterns. Secured 3rd place in a Kaggle hackathon by enabling targeted retention strategies based on customer risk scores.

🏠 Housing Price Prediction – Ridge & Lasso Regression

Repo: House Price Prediction
Models Used: Linear Regression, Ridge Regression, Lasso Regression
Description: A regression-based predictive model using regularization (Ridge & Lasso) to estimate Australian housing prices for investment decisions. Implements robust feature selection and tuning for generalizability and business insight.

Repo: Bike Sharing System
Models Used: Linear Regression
Description: A predictive analytics project for BoomBikes to estimate daily shared bike demand using weather, seasonality, and usage data. Helps drive post-COVID business strategy by identifying key demand factors through regression modeling.

💳 Lending Club Loan Default Analysis – Risk Insights with EDA

Repo: Lending Club Loan Data Analysis
Models Used: Exploratory Data Analysis (EDA)
Description: A deep dive into Lending Club’s loan dataset to uncover patterns in loan default behavior using EDA. Supports risk mitigation by highlighting correlations with interest rates, loan purpose, borrower income, and geographic distribution.

📫 Contact

Visitor Count

Thanks for visiting! Feel free to explore these projects and reach out for collaboration or discussion. 🙌