Summary
Overview
Work History
Education
Skills
Websites
Certification
Accomplishments
Awards
Timeline
Generic

Naga Venkata Aakash Devavarapu

Senior Machine Learning Engineer
Austin,TX

Summary

Motivated and highly collaborative ML Engineer with a strong background in object-oriented design and developing scalable AI solutions. Reduced operating costs by 50% and boosted software sales by 40%. Proficient in collaborating with cross-functional teams to make data-driven decisions using LLM performance analytics, resulting in a 20% monthly organic user growth.

Overview

6
6
years of professional experience
1
1
Certification
3
3
Languages

Work History

Senior Machine Learning Engineer, Generative AI

Publishing.com
Austin, Texas
09.2024 - 04.2025
  • Proposed, designed and deployed new AI architecture for manuscript writing using agents and complex LLM flows using LangGraph, LangChain on AWS saving $25,000 monthly ensuring scalability to 45,000 daily active users.
  • Improved Manuscript quality by upgrading LLM workflows to complex multi-agent systems adding self-evaluation and correction tools with LangGraph driving software sales by 40%.
  • Developed recommendation system using Agents with reasoning and market research tools to recommend personalized book topics, increasing author monetization rates by 20%.
  • Optimized prompts across all LLM services with prompt engineering best practices, cutting down unwanted examples, strengthening instructions and guard rails resulting in 25% token cost reduction.
  • Created custom evaluation pipelines in MLFlow with statistical and LLM-as-Judge metrics specific to long text, benchmarking against ground values resulting in a 30% reduction in quality complaints.
  • Integrated content moderation in topic generation and final draft phase of book writing with sentiment analysis and LLM-based grounding with 94% Recall score.
  • Fine-tuned Flux.1 with Dreambooth LoRA to generate book covers for users based on title and style deployed to EC2 instance as a container as new feature resulting in addition of 2000 new paid users.
  • Drove collaboration with Product team to develop custom dashboards visualizing LLM performance over time during A/B tests, resulting in a 15% increase in subscription upgrades to highest plan.

Machine Learning Engineer, Healthcare Industry

Hexaware Technologies
Bethesda, Maryland
01.2023 - 09.2024
  • Led project for a healthcare provider that optimized claim processing by 50% through direct extraction of medical codes from patient clinical notes, adhering to HIPAA compliance with $2M budget and 5 direct reports.
  • Improved reliability by fine-tuning Mistral-7B with LoRA on multi-GPU EC2 instance in AWS increasing Recall score to 93%.
  • Developed a RAG system using Mistral-7B and PineconeDB for semantic extraction of ICD codes, incorporating hierarchical mapping and synonym generation improving entity recognition accuracy to 91%.
  • Efficient communication with leadership to convert business requirements into tech specs along with project planning and explaining technical concepts to non-technical product teams, resulting in a 4-week early release following agile best practices.
  • Implemented ETL data pipelines for preprocessing of 100,000 text documents with compliance verification, document parsing, PII removal, tokenization and other NLP cleaning using PySpark with AWS Glue for training models in SageMaker.
  • Performed data processing and EDA on unstructured documents applying imputation by combining BERT and KNN for missing sections, improving overall model accuracy by 10%.
  • Implemented multi-step LLM uncertainty quantification with token confidence, vector similarity and self-evaluation methods to flag inconsistent extractions with human-in-the-loop review, resulting in 5% increase in recall score.

Machine Learning Researcher

University of Florida Innovation Hub
03.2022 - 01.2023
  • Achieved $1.5M in funding by effectively demonstrating AI solutions as founding ML Engineer for a startup, resulting in $250,000 in annual revenue during the pilot phase.
  • Trained a Computer Vision model with CNN+Vision Transformer in PyTorch to analyze wound images and predict healing time with 0.17 MAE and deployed to AWS Sagemaker with serverless configuration for cost efficiency.
  • Developed GANs to generate synthetic wound images, increasing dataset size from 125 to 700 and tuned model parameters with FID, SSIM and MSE metrics with Weights & Biases.
  • Finetuned YOLOv4 to extract general notes from document images fed to Tesseract OCR Engine to extract text with 91% accuracy on 14,000 high resolution floor plan PDFs.
  • Improved prediction accuracy from initial dataset at 77% to 84% by adding more images and feature engineering to include new features like depth, color degradation, area, location of wound, etc with image EXIF and OpenCV library.
  • Created end to end inference pipeline with AWS Step Functions and Lambda for image upload, preprocessing, prediction delivery serving ~3000 requests/day.
  • Developed 2-stage verification system using semantic ranking with RoBERTa embeddings and fine-tuned BERT classifier to verify if the building floor plan is compliant with FBCR regulations with an 90% precision score.
  • Developed and implemented time series modeling with ARIMA and spectral analysis to identify and cluster recurring patterns in EEG data, improving abnormal activity detection accuracy by 15%.

Natural Language Processing (NLP) Engineer

Deloitte US
01.2019 - 12.2021
  • Increased chatbot accuracy from 90% to 95% by tuning hyperparameters with Azure HypeDrive and increasing dataset quality in Azure Machine Learning Datasets for Enterprise.
  • Added question generation capabilities during chatbot training by fine-tuning BERT and T5 transformers with HuggingFace to generate training data suggestions using PyTorch reducing manual data creation time by 40%.
  • Developed NLU chatbots for enterprise applications using Microsoft LUIS in Azure cloud maintaining average accuracy of 90% across 17 chatbots for multiple business teams.
  • Used AzureML to create CI/CD deployment pipeline to collect new data from user feedback real-time and retrain model, enabling continuous improvement of model performance reducing delivery time by 60%.
  • Recipient of "Applause Award" for introducing new ideas involving AI and machine learning methods that improved product delivery time by 30%.

Education

Master of Science - Computer Science, Machine Learning

University of Florida
Gainesville, Florida
01.2022 - 05.2023

Bachelor of Engineering - Electrical, Electronics And Communications Engineering

Osmania University
Hyderabad, India
05.2001 -

Skills

  • AWS

  • Deep Learning

  • Docker

  • LLMs

  • NLP

  • NLTK

  • Python

  • PyTorch

  • SpaCy

  • TensorFlow

Certification

TensorFlow Certified Developer, Google TensorFlow

Accomplishments

  • Supervised team of 5 offshore staff members following Agile methodology.
  • Achieved $50,000 in savings by introducing LangSmith for prompt engineering by adding GitHub like actions and approvals organization wide.
  • Collaborated with team of 3 Sports Physicians in the development of Wound Recovery Prediction explaining ML models to doctors.
  • Achieved 40% improvement in chatbot delivery time with BERT and T5 transformers.

Awards

  • Applause Award for introducing Generative AI applications in chatbot development at Deloitte.
  • Received Silver Medal from Brandon Hall Group Excellence Awards for "Best Advance in AI and Machine Learning" in Learning and Development Category in Deloitte.

Timeline

Senior Machine Learning Engineer, Generative AI

Publishing.com
09.2024 - 04.2025

Machine Learning Engineer, Healthcare Industry

Hexaware Technologies
01.2023 - 09.2024

TensorFlow Certified Developer, Google TensorFlow

11-2022

Machine Learning Researcher

University of Florida Innovation Hub
03.2022 - 01.2023

Master of Science - Computer Science, Machine Learning

University of Florida
01.2022 - 05.2023

Natural Language Processing (NLP) Engineer

Deloitte US
01.2019 - 12.2021

Bachelor of Engineering - Electrical, Electronics And Communications Engineering

Osmania University
05.2001 -

Similar Profiles

CREATE PROFILE
Naga Venkata Aakash DevavarapuSenior Machine Learning Engineer