Headline

AI Engineer | Building Production Grade Multi AI Agent & RAG Systems | LangChain, CrewAI, FastAPI, Pinecone | AWS Certified AI Practitioner | Gold Medalist @SoCSE KLETECH |

About

I'm an AI Engineer who builds intelligent systems which includes RAG pipelines, multi agent workflows, and full stack AI apps that actually solve problems. At Outamation, I built a Document Retrieval System for 100+ page mortgage documents using LlamaIndex and Pinecone hybrid search. Combined dense and sparse vectors with cross encoder reranking, pushed Recall@6 from 88.8% to 96.4%, and cut document query time from 10 minutes to 10 seconds. Evaluated the pipeline with RAGAS framework across 500+ queries achieving 84% Answer Correctness, 87% Context Precision. On the agentic side, I designed a multi crew pipeline with 5 specialized CrewAI agents for lead scoring and qualification, hitting a 9.7/10 average task quality score. Also built an E Commerce AI Agent deployed on Render that routes queries between Products and FAQ data with 97.4% accuracy using Gemini function calling. Before GenAI, I worked on ML at Pravinya IT Services where I optimized health insurance premium prediction models, selected XGBoost over Random Forest, brought RMSE down from Rs. 2500 to Rs. 1300, and improved R² from 0.84 to 0.88 using Optuna. I work primarily with Python, LlamaIndex, CrewAI, FastAPI, Pinecone, and React. My core focus is on designing and scaling AI systems that work in production, the framework or tool can change, but the ability to build reliable, end to end solutions stays the same. Currently looking for GenAI and AI/ML Engineering roles where I can design and deploy intelligent systems at scale.

Location

Hubli

AI stack

codex

opencode

Work experience

Extern, Outamation at Extern, Dec 2025 - Feb 2026 · 3 mos, Hubli, Developed a Document retrieval system for mortgage documents, reducing query time and improving information access accuracy. 1. Built a RAG pipeline that automated information extraction from 100+ page mortgage documents, reducing document query time from 10 minutes to 10 seconds. 2. Engineered Hybrid RAG + Cross Encoder Reranking by combining dense and sparse vectors through Pinecone hybrid search, improving Recall@6 from 88.8% to 96.4%. 3. Implemented RAGAS evaluation framework to evaluate RAG pipeline across 500+ queries, achieving an Answer Correctness of 84% and Context Precision of 87% across the evaluation suite. 4. Integrated the system using LlamaIndex and Python, enabling non-technical users to quickly find specific information within complex financial documents.

Machine Learning Intern at Pravinya Information Technology Services, Apr 2025 - Oct 2025 · 7 mos, Hubli, Karnataka, India, Built and optimized predictive models for health insurance premium price prediction, improving accuracy and delivering actionable insights to stakeholders. 1. Enhanced model interpretability by reducing feature set by 20% through multicollinearity analysis (VIF), enabling clearer insights for decision-making. 2. Optimized model performance by selecting XGBoost over Random Forest, reducing RMSE from Rs. 2500 to Rs. 1300. Implemented Optuna for hyperparameter optimization, improving R² score from 0.84 to 0.88. 3. Conducted root cause analysis on model predictions, identifying that 30% of high-error cases stemmed from insufficient data in the young age segment, and recommended targeted data collection strategies. 4. Presented model findings and recommendations to client stakeholders through an interactive Streamlit application, translating technical insights into actionable business strategies.

Research Experience for Undergraduates (REU) at KLE Technological University - Hubballi (India), Aug 2023 - Dec 2023 · 5 mos, Hubli, Designed a novel blockchain consensus algorithm combining game theory with distributed systems to improve Ethereum's security and efficiency. 1. Proposed a two-stage game theory-based consensus algorithm (DDPoS) that addresses Ethereum blockchain limitations including centralization risks and inefficient node downgrading. 2. Engineered and tested the algorithm on an Ethereum testbed with 25 nodes and 500 transactions, achieving a 5.92% increase in fairness and 31.7% improvement in malicious node detection. 3. Simulated large-scale deployment across 1,000 nodes and 50,000 transactions, reducing transaction time by 27.04% and increasing block creation efficiency by 11.26% compared to existing algorithms.

Education

Bachelor's degree, Computer Science and Engineering, KLE Technological University - Hubballi (India)