RPRavivarma Patturi
AI/ML Engineer · 4 Years Experience · Bengaluru, India

Ravivarma Patturi

AI Engineer: production agentic systems & edge inference

I design and deploy production-grade Generative AI, multi-agent decision systems, and high-throughput inference pipelines for leading U.S. healthcare payers and providers. Alongside enterprise delivery, I follow on-device AI out of personal interest and build in that space: InferEdge AI (on-device speech-to-speech with VaayuEdge), Neural Mastery (1,800+ page AI engineering platform), published Hugging Face models, and technical teaching.

Hire / ContactDownload résumé
Direct:patturiravivarma@gmail.com/LinkedIn/GitHub
Ravivarma Patturi, AI and Machine Learning Engineer
Current: AI Engineer @ Autonomize AI
14 Days
PoC to Production

Delivered healthcare benefits digitization PoC in 14 days, converting directly into a full-scale enterprise production engagement.

80%
Lower Adjudication Cost

Built multi-agent SOP reasoning system that automates claims adjudication against business rules with explainable audit trails.

70,000+
Healthcare Concepts Indexed

Clinical AI semantic search pipeline with Azure AI Search and cross-encoder reranking for natural-language expression extraction.

2× Faster
50% Smaller at Edge

Quantized deep sequential models with <1.2% MAPE deviation, enabling real-time edge battery management deployment.

01/Work History

Production Experience

Demonstrated track record of taking complex AI systems from research and rapid proof-of-concept into hardened, high-availability enterprise production.

AI Engineer

Autonomize AI
Feb 2026 – PresentBengaluru, India

Architecting enterprise agentic platforms, knowledge graphs, and clinical reasoning systems for leading U.S. healthcare payers and providers.

Recursive Language Model & Agent Platform:Designed and implemented the Recursive Language Model (RLM) architecture enabling hierarchical reasoning through Master–Sub LLM collaboration; built the orchestration layer on Temporal for scalable, fault-tolerant execution alongside ReAct and Plan-and-Execute workflows.
Healthcare Benefits Digitization (PoC to Production):Delivered an automated benefit-extraction PoC for a leading U.S. healthcare payer in 14 days, converting it into a full-scale production system; engineered LLM + Knowledge Graph extraction pipelines adhering to ICD and CPT coding standards.
Multi-Agent Claims Adjudication (80% Cost Reduction):Built a multi-agent system reasoning over complex SOP documents that reduced claims adjudication cost by 80%; engineered a Knowledge Graph Generation Agent to convert policy rules into structured decision trees and an Adjudication Agent producing transparent, auditable Approve/Reject/Pend reasoning.
Clinical Search & Inquiry Classification:Indexed 70,000+ healthcare concepts in Azure AI Search with cross-encoder reranking to convert natural-language queries into logical campaign expressions; developed a ModernBERT classifier handling class imbalance via synthetic generation and oversampling.
LangGraphTemporalKnowledge GraphsAzure AI SearchModernBERTCross-EncoderFastAPIPythonICD/CPT

AI Research Engineer

Boltzmann Labs Pvt Ltd(Previously Deep Learning Research Intern (Nov 2022 – Jun 2023))
Jul 2023 – Jan 2026Bengaluru, India

Led research and production deployment of agentic hardware verification workflows, domain-specific LLM fine-tuning, and edge inference optimization.

Agentic Workflow for RTL Verification:Built an LLM agentic workflow that extracts functional intent from IC specifications and generates coverage-aligned test plans with targeted stimulus; integrated directly into Cadence Xcelium UVM flows, significantly accelerating coverage closure.
Edge Inference Optimization for Battery Management:Trained PyTorch sequential models (RNN/LSTM/GRU) achieving 99% reliability in state-of-charge prediction; applied dynamic quantization achieving 50% model size reduction and 2× lower latency with <1.2% MAPE deviation for real-time edge deployment.
Domain LLM Fine-Tuning & Specialized RAG:Fine-tuned LLaMA-3.2-1B with LoRA on NVIDIA L40S GPUs for chemically valid crystal structure generation; developed a domain-specific RAG system (LangChain, Mistral API, ChromaDB) for context-aware retrieval across heterogeneous battery research data.
GPU Microservices & Molecular AI Pipelines:Engineered a GPU-accelerated, Dockerized GROMACS + OpenMM simulation workflow deployed as a FastAPI microservice on AWS EKS (40% faster setup, 30% higher pipeline efficiency); contributed multi-objective ML optimization to Boltchem small-molecule discovery.
PyTorchLLaMA-3.2 (LoRA)NVIDIA L40SQuantizationCadence XceliumFastAPIDockerAWS EKSChromaDB
02/Independent Work

What I'm Building

Beyond enterprise engineering, I build in public across edge inference, technical platforms, open research models, and engineering education.

On-Device Inference

InferEdge AI · VaayuEdge

Building ultra-efficient on-device multimodal inference; VaayuEdge runs English→Telugu speech translation client-side in the browser with zero cloud roundtrips.

Try VaayuEdge Demo
Interactive Learning Platform

Neural Mastery

Self-built AI engineering platform featuring 1,800+ pages of deep learning, GenAI, and agentic curricula with in-browser Python execution and visual labs.

Explore Platform
Research Checkpoints

Hugging Face Models

Published fine-tuned IndicTrans2 English→Telugu research checkpoints optimized for specialized domain vocabulary and low-resource dialect translation.

View Checkpoints
Technical Education

Neural Mastery on YouTube

Companion video series breaking down machine learning foundations, backpropagation, and agentic workflows in Telugu and English for an engaged engineering community.

Watch on YouTube
03/Code & Architecture

Selected Projects

Four flagship open-source architectures demonstrating agent orchestration, advanced retrieval, edge-ready ML pipelines, and interactive systems.

Enterprise RAG

Regulatory QA RAG Chatbot

qa_rag_application
Problem:

Regulatory compliance and policy documents are dense and complex. Generic search tools often return fragmented clauses or hallucinated answers, creating significant compliance risks.

Approach:

Architected a production-ready conversational RAG assistant using GPT-4o. Combined dense vector search with sparse BM25 retrieval (hybrid retrieval), query rewriting and memory-augmented context. Fully containerized with Docker for Kubernetes deployment.

Result:

A containerized assistant, deployed on Kubernetes, that answers questions over regulatory documents from retrieved passages instead of from the model's memory.

PythonGPT-4oLangChainHybrid RetrievalDockerKubernetesVector DB
View Repository
Multi-Agent System

Financial Performance Reporting Agent

Finance_Agent
Problem:

Synthesizing quarterly financial disclosures, SEC filings, and competitor metrics is an exhaustive manual undertaking that delays strategic investment reporting.

Approach:

Designed a multi-agent workflow orchestrated with LangGraph and OpenAI. Implemented specialized agents for financial statement extraction, web-scale competitive benchmarking via Tavily Search, and structured narrative generation, wrapped in a Streamlit UI with human-in-the-loop review.

Result:

Turns an uploaded CSV of company financials into a narrative report with competitor benchmarks, refined through user feedback in a Streamlit UI.

LangGraphOpenAITavily SearchStreamlitPandasPython
View Repository
Top 10 MachineHack 2024

Municipal Traffic Volume Forecasting

Traffic_Volume_Prediction_Hackathon
Problem:

Predicting municipal traffic congestion requires capturing complex cyclical patterns influenced by severe weather changes, irregular holiday calendars, and sudden peak events.

Approach:

Engineered high-signal spatio-temporal features, cyclic time encodings, and weather interaction variables. Developed an ensemble XGBoost model tuned with stratified cross-validation and hyperparameter optimization.

Result:

Placed in the Top 10 of the MachineHack Great Indian Hackathon 2024.

XGBoostScikit-LearnFeature EngineeringPandasNumPy
View Repository
Learning Platform & OSS

Neural Mastery Platform Engine

NeuralMastery
Problem:

Most AI learners face friction between theoretical lectures and practical code verification, lacking immediate in-browser execution or visual intuition.

Approach:

Built a modern interactive education platform featuring 1,800+ curriculum pages, in-browser Python execution via Pyodide (WebAssembly), interactive visual labs, progress tracking and a community leaderboard.

Result:

A live, free platform at neuralmasteryai.com with interactive visualizations, coding labs, interview practice and a companion YouTube channel.

ReactTypeScriptVitePyodideMDX
View Repository
04/Education & Media

Writing & Videos

Demystifying AI/ML foundations, agentic engineering, and deep learning architectures through structured platform writing and video tutorials.

Interactive Curricula

Neural Mastery Platform & Engineering Blog

Over 1,800+ pages of deep engineering guides covering neural network internals, large language model fine-tuning with LoRA, agentic orchestration, and client-side browser inference with interactive Python execution.

Visit NeuralMasteryAI.comSubscribe on YouTube (@neuralmastery)

Featured Video Deep Dives

Browse all videos on YouTube

Introduction to Evaluation Metrics in Telugu | Machine Learning

Deep dive into classification metrics — Precision, Recall, F1-Score, ROC-AUC, and confusion matrix interpretation for production ML systems.

Backward Propagation in Telugu | Deep Learning

Step-by-step mathematical breakdown and derivation of the chain rule, gradient updates, and backpropagation in deep neural networks.

Normalization vs Standardization in Telugu | Machine Learning

Clear architectural guidance on feature scaling techniques, mathematical formulations, and when to use MinMax vs StandardScaler.

05/Capabilities

Technical Skills

A battle-tested stack spanning multi-agent orchestration, transformer fine-tuning, edge quantization, and production cloud infrastructure.

GenAI & Agentic Systems

Production agent architectures, multi-agent coordination, structured reasoning, and retrieval.

Large Language Models (LLMs)Multi-Agent SystemsLangGraph & LangChainLlamaIndexRecursive Language Models (RLM)ReAct & Plan-and-ExecuteHybrid RAG (Dense + BM25)Cross-Encoder RerankingQuery RewritingKnowledge GraphsFine-Tuning (LoRA)Prompt Engineering

Machine Learning & Deep Learning

Neural network architecture design, representation learning, and domain-specific modeling.

PyTorchTensorFlowHugging Face TransformersModernBERTRNN / LSTM / GRUCNNs & AutoencodersXGBoostscikit-learnNLTK & spaCyPhysics-Informed SurrogatesMulti-Objective OptimizationHyperparameter Tuning

Inference & Edge AI

Quantization, latency minimization, and client-side web inference execution.

Dynamic Quantization (INT8)ONNX RuntimeIn-Browser Speech & AudioWebAssembly / PyodideIndicTrans2 TranslationEdge Model BenchmarkingNVIDIA L40S GPU Workflows

Backend & MLOps Infrastructure

Distributed workflow orchestration, containerization, search indexes, and cloud APIs.

Python (Core & Async)FastAPI MicroservicesTemporal Workflow OrchestrationDocker & ContainerizationKubernetes (AWS EKS)Azure AI SearchChromaDB & Vector StoresAWS & Cloud ServicesStreamlit UIGit & CI/CD ActionsLinux Systems
06/Get in Touch

Let's build together.

I am always interested in discussing production agentic systems, on-device edge AI, healthcare AI workflows, or high-impact engineering opportunities.

Direct Contactpatturiravivarma@gmail.com
Professional Profilelinkedin.com/in/ravivarma-patturi
Open Source Codegithub.com/ravivarmapatturi
Building in Public@NeuralMastery
Looking for a comprehensive PDF summary?Updated October 2026 with production metrics and engineering outcomes.
Download Résumé (PDF)