Shivam Modi

Built and delivered production-ready AI systems end-to-end, including a full-stack RAG-based insurance claim verification platform and a self-correcting Text-to-SQL agent using FastAPI, LangChain, Docker, and MySQL. The claim verification system handled policy PDFs, uploaded bills, and exclusion logic, while the text-to-SQL system included SQL guardrails, agentic workflow design, and achieved 0.86+ faithfulness in RAGAS evaluation. At Oracle, I contributed to 25+ component upgrades and validated 40+ patches across multiple branches, debugged 50+ build/compiler issues on Windows and Linux, automated maintenance workflows with Python, and wrote 25+ technical wikis while mentoring 4 interns to improve onboarding and knowledge transfer.

Lead Gen AI Engineer at ClaimTrackr (2026-03 – Present)

Full-Stack RAG System for Insurance Claim Verification

• Architected an end-to-end RAG system for insurance claim evaluation across 5 policy PDFs, uploaded bills data, and 7 exclusion categories using FastAPI, Pinecone, and SentenceTransformers
• Delivered a production-ready full-stack system using React + Vite + Tailwind, backend APIs, and Docker Compose for reproducible deployment
• Executed a retrieval + reasoning pipeline combining semantic search with Gemini LLM to control model invocation, extract diseases/expenses, validate claims, and generate explainable verdicts
• Optimized Docker build using multi-stage setup, reducing image size from 919MB to 455MB (≈50%)

LLM Systems Engineer | Text-to-SQL AI System (2026-02 – Present)

Natural Language to SQL with Self-Correction & Agentic Workflow

• Built a natural-language-to-SQL system using FastAPI, Streamlit, LangChain, Gemini/Groq, and MySQL 8.0 to generate executable queries from user questions
• Implemented SQL guardrails and automatic self-correction to block unsafe queries and repair failed SQL
• Evaluated system performance using RAGAS (0.86+ faithfulness) and optimized Docker deployment with a multi-stage build, reducing image size from 2.1GB to 516MB (≈75%)
• Designed a LangGraph + ReAct agentic workflow for schema-aware SQL generation

Member Technical Staff at Oracle India Private Limited (2023-06 – 2025-09)

• Automated maintenance workflows using Python scripts, reducing manual effort for updates & deployment
• Delivered 25+ component upgrades (OpenSSL, PCRE2, ModSecurity, curl, zlib) and validated 40+ patches across 5 OHS branches, improving release stability and reducing regression risk in quarterly builds
• Debugged 50+ build and compiler issues across Windows & Linux, including resolving a CVSS 7.8 vulnerability via Intel ICC upgrade, improving build reliability and binary integration
• Authored 25+ technical wikis and mentored 4 interns, reducing onboarding time by 60% and improving knowledge transfer across development workflows

Machine Learning Engineer at Oracle India Private Limited (2022-05 – 2022-07)

• Deployed a real-time anomaly detection pipeline using Oracle AI and deep learning (LSTM/RNN) to predict US-region server downtime from load metrics
• Benchmarked Oracle AI vs traditional ML models, achieving 11% faster detection over baseline

Data Scientist at Ultratech Cement (2021-06 – 2021-07)

• Mechanized a Scikit-learn regression model for lead conversion propensity on 1.2M+ records, achieving 91% accuracy, enabling automated lead scoring and improving marketing decision-making
• Built an image classification pipeline using TensorFlow/Keras on 1000+ images to enable automated meta-tagging across 24 features

Bachelor of Technology in Electrical Engineering – Indian Institute of Technology Bombay (2019-07 – 2023-05)