Generative AI Engineer at Carelon Global Solutions

Ravuri Sai Vidyasagar

▌

I build LLM systems that hold up in production — RAG pipelines, agentic workflows, and document intelligence for healthcare.

Selected Work Get in Touch
40% Faster healthcare workflows via LLM-based automation of prior authorization
5,000+ Daily active users served across production AI systems
1,000+ Daily transactions patient case submissions at peak
99.5% Production uptime on patient-facing ML systems
01

About

I'm a Generative AI Engineer with five years spent taking LLM systems from prototype to production — mostly in healthcare, where the margin for error is thin and every output needs an audit trail.

My work sits at the intersection of retrieval, reasoning and document understanding: RAG architectures grounded in payer guidelines, agentic pipelines that make prior-authorization decisions, and OCR workflows that turn scanned prescriptions and clinical notes into structured data. I care about systems that are explainable and observable, not just accurate on a benchmark.

Today I'm at Carelon Global Solutions (Elevance Health), building agents for US healthcare — benefit leakage detection, benefits trend analysis and prior authorization. Before that I architected LLM and RAG workflows for healthcare payers at PenguinAI India, and built a no-code ML platform used by 200+ people at Awone AI.

sagarravuri2@gmail.com LinkedIn ↗ Download résumé ↓

Applied LLM Systems

Fine-tuning, prompting and evaluating GPT-4, LLaMA 3 and Qwen for production workloads.

RAG & Retrieval

Grounding model output in domain corpora with Pinecone, FAISS and hybrid retrieval strategies.

Agentic Pipelines

Multi-step reasoning agents with tool use, validation gates and transparent audit trails.

Document Intelligence

OCR and vision-language models that convert unstructured clinical documents into structured data.

02

Selected Work

What I'm building now, and the systems that shipped before it.

In progress

Carelon Global Solutions (Elevance Health)

Agents I'm building now for US healthcare payer operations.

Benefit Leakage Detection

Agents that surface where benefits are being paid out incorrectly, flagging leakage patterns across claims and benefit configurations for payer review.

Agentic AI LLMs Claims Data

Benefits Trends

Agents that analyse benefit utilisation across populations, turning claims and benefits data into trends decision-makers can act on.

Agentic AI LLMs Analytics

Prior Authorization

Agentic review of authorization requests against clinical policy, continuing prior-auth automation work at payer scale.

Agentic AI RAG Clinical Policy

Shipped

Healthcare · RAG

HCS Healthcare

PenguinAI India 2024 — 2026

A guideline-aware RAG agent for prior authorization

The problem

Prior authorization reviews required staff to manually cross-reference each request against payer-specific clinical guidelines — slow, inconsistent, and hard to scale across multiple payers.

The approach

Built a Qwen-based dynamic questionnaire engine backed by a Pinecone vector database, retrieving the relevant guideline passages in real time so each generated question is grounded in the payer's own policy rather than the model's priors.

40% less manual PA processing
25% faster approval turnaround
95%+ retrieval accuracy
500+ daily authorization requests
Qwen RAG Pinecone LangChain Python
Healthcare · Agentic AI

Snowflake PA

PenguinAI India 2024 — 2026

An agentic decision engine with a full audit trail

The problem

Authorization decisions depended on reading scanned documents, mapping them to ICD-10/CPT codes, and checking eligibility — a chain of judgement calls that regulators expect to be traceable end to end.

The approach

Designed an agentic pipeline chaining OCR, code mapping and eligibility validation, each step surfacing its reasoning to a transparent audit UI. Shipped as a FastAPI microservice backed by MongoDB.

30% faster authorization decisions
300+ concurrent API requests
<1s response time
Agentic AI FastAPI MongoDB OCR ICD-10/CPT
Healthcare · RAG + ML

Ovation Billing Review

PenguinAI India 2024 — 2026

Explainable denial prediction on 50,000+ claims

The problem

Billing teams discovered claim denials only after submission, and a black-box risk score would not have been actionable — reviewers needed to know which features drove each prediction.

The approach

Paired a RAG-based billing agent that extracts CPT/ICD codes into structured YAML with an XGBoost denial-prediction model served over FastAPI, using SHAP to expose per-claim feature attribution to reviewers.

87% model precision
50,000+ historical claims engineered
RAG XGBoost SHAP FastAPI YAML

Also built

Respire — Prescription Processing

OCR + NLP workflow extracting demographics, diagnosis codes and nutritional plans from prescriptions at 98% accuracy, saving 12+ hours of manual transcription weekly.

OCR NLP Document AI

CIBI — No-Code AI Platform

End-to-end no-code interface for ML/DL model training, evaluation and deployment, cutting development time by 70% for 200+ non-engineering users.

MLOps FastAPI No-Code

Government Maps Digitization

OCR framework extracting numeric depth values from bathymetric maps at ±50m accuracy, halving manual digitization time for marine authority workflows.

OCR OpenCV Geospatial

Respire DPP — Patient Risk Scoring

Risk stratification model categorizing patients into nutritional risk tiers and generating personalized dietary recommendations.

Risk Modeling Healthcare ML
03

Try It

Three working demos of the patterns behind the systems above — run them yourself.

Before you start

These run deterministic local logic over synthetic guideline and claims data — they illustrate the approach, not the production systems, and no real patient data is involved. Please don't enter real patient information.

Retrieval over payer guidelines

Ask a coverage question. The query is expanded with clinical synonyms, scored against all 10 synthetic guideline passages by TF-IDF cosine similarity, and the top matches come back with their scores — the grounding step that keeps a RAG answer tied to the payer's own policy instead of the model's priors.

Try

Ranked passages will appear here.

04

Experience

Generative AI Engineer

Carelon Global Solutions (Elevance Health) Current

Building agentic AI systems for US healthcare payer operations — benefit leakage detection, benefits trend analysis and prior authorization.

Jun 2026 — Present Hyderabad

Generative AI Engineer

PenguinAI India

Architected end-to-end AI workflows — LLMs, RAG, OCR and agent-based reasoning — for healthcare payers and providers.

Jul 2024 — Mar 2026 Hyderabad

AI Engineer

Awone AI

Built a Python ML toolkit and no-code platform that let 200+ users ship production-grade models with minimal code.

Mar 2023 — Jun 2024 Hyderabad

AI & Software Engineer

Pyro Holdings

Orchestrated data pipelines and automated reporting in Python and SQL, improving turnaround by 35%.

Aug 2021 — Feb 2023 Hyderabad
Education

B.E. Computer Science & Engineering

Sathyabama Institute of Science and Technology · Chennai

2017 — 2021
05

Capabilities

What I reach for, and why.

LLMs & Fine-Tuning

Adapting foundation models to domain tasks and evaluating them beyond benchmark accuracy.

GPT-4 LLaMA 3 Qwen BERT RLHF Prompt Engineering

RAG & Vector Search

Retrieval architectures that keep generated output grounded in source documents.

LangChain Pinecone FAISS Weaviate Embeddings Semantic Search

Document Intelligence

Turning scanned, unstructured clinical documents into reliable structured data.

LayoutLM OpenCV Vision-Language Models Multi-modal AI

AI Infrastructure

Serving models as observable, well-behaved services that survive real traffic.

FastAPI Flask Docker MongoDB REST APIs Microservices

ML Engineering

Classical ML where it outperforms an LLM, with interpretability built in.

PyTorch HuggingFace Scikit-learn XGBoost SHAP

Healthcare AI

Domain workflows where compliance and auditability are requirements, not extras.

Prior Authorization Clinical NLP ICD-10/CPT EHR Integration HIPAA
06

Get in Touch

Open to Generative AI and LLM engineering roles, and to interesting problems in applied AI.

Email sagarravuri2@gmail.com Phone +91-7993425279 LinkedIn Connect ↗
Location Hyderabad, India