Shiva Sankeerth Reddy Yarradla Email & Phone Number

San Jose, Ca, Us

Palo Alto, California, Us

Working with ex-Googlers & the former founders to develop an AI product for automating finance workflows using advanced AI agents.- First Hire in the AI/ML Team.- Zero-to-One Product Development: Rapidly prototyped, evaluated, and integrated new product features, driving innovative solutions from inception to implementation.- Developed and continuously improved best-in-class LLM methods for zero-shot dense retrieval, large-scale hybrid semantic search, and recursive abstractive document analysis.- Agentic Workflows Design: Designed and implemented LLM agents that utilized advanced primitives and external tools to deliver high-quality, hallucination-free, long-form work outputs efficiently.- Established rigorous LLM evaluation protocols, creating custom metrics and combining human preference judgments from domain experts with synthetic data generation for comprehensive assessment.- Fine-tuned LLMs, Cross Encoders (as Rerankers) and domain-specific embeddings models tailored for specialized knowledge work tasks, enhancing performance and relevance.The product would go on to secure millions in ARR with some of the top investment firms across the Globe.

Stony Brook, Ny, Us

Graded assignments and projects, providing detailed feedback to students on their assignments.Provided one-on-one assistance to students during office hours, answering questions and offering support with SQL and database management-related projects and assignmentsCoached students on SQL syntax, query optimization, and debugging techniques during weekly office hours, helping them to develop strong database management skills

San Jose, Ca, Us

Member of the "Contextual Understanding and Predictive Analytics" team, using ML and NLP to provide a world-class Conversation intelligence platform to billions of Zoom users. • Constructed an NLP pipeline to extract transcripts from live video, predicting speaker confidence for a conversational intelligence platform by researching audio, video, and transcript data features.• Modeled the problem as a regression and classification task, developed a custom sklearn pipeline with multiple BERT and Sentence Transformer SBERT variants as feature extractors, improving final predictions by 32%.• Applied distributed training techniques using SageMaker, reducing convergence time by 68%, and deploying final models with 10.4% improvement in F1-score for classification and 20% improvement in RMSE for regression tasks.• Deployed the final model as a micro-service using Flask and Kubernetes on Triton GPU Inference Server, applying quantization and knowledge distillation techniques, reducing inference time by 75%.

Stony Brook, Ny, Us

Working in the Pervasive Computing and Smart Sensing (PiCASSo) Lab under the guidance of Prof. Shubham Jain. The lab mainly focuses on leveraging multimodal sensing devices for addressing challenges in the space of health and urban communities.• Working with Prof. Shubham Jain at Mobile Systems Lab to find the anomalies in the time series data obtained from mobile and smart-watch sensors by using a combination of Stacked and Contractive Autoencoders.• Analyzed the limitations of Stacked and convolutional Autoencoders in learning temporal and spatial frequency variations.• An Autoencoder with convolutional Encoding and Stacked Decoding and a pre-processing technique of Dynamic Time warping with time series clustering were proposed to improve the anomaly detection rate from 54% to 95%.

Stony Brook, Ny, Us

Worked as a Graduate Teaching Assistant for CSE 114: Object-Oriented Programming under the guidance of Prof. Praveen Tripathi.Responsibilities:• Taught OOPs concepts using Java Programming Language to a class of 35 undergraduate students and also gave hands-on tutorials during 30+ Lab sessions.• Graded the bi-weekly homework assignments and weekly Lab assignments.• Conducted 50+ office hours for resolving doubts in class content and assignments.

Hyderabad, In

Project-1: Designing and Developing a end-to-end chatbot application• Spearheaded a team of 4 peers to develop an in-house Chatbot UI kit for integration with Android.• Designed and developed the pipeline of a virtual assistant chatbot capable of engaging in dialogues and question-answering from textual/tabular documents using RASA and BERT-QA and improved the accuracy from 65% to 95% for classifying long documents into multiple classes using hyperparameter tuning.• Enhanced the API response time by 25% through remodeling Database Schema.• Dockerized all micro-service modules to deploy on Kubernetes for a 99% system uptime.Project-2: Designing a Data pipeline for Healthcare and IoT clients.• Utilized multiple AWS services like Lambda, SQS, SNS, Cloud Watch, EC2, and EKS for deploying the system.• Maintained Amazon EMR clusters for handling spark and running ETL jobs. Also handled scaling with job-specific configurations.• Implemented Apache Airflow in the ETL process for scheduling and monitoring jobs. Worked on customizing operators and callbacks.• Built AWS QuickSight reports to monitor operational KPIs SLAs like average DAG runtimes and successful dispatches to ensure high availability and quick resolution of customer issues.• Designed and implemented an AWS Glue metadata architecture to automate pipelines between AWS Glue and Amazon S3 for ETL operations, resulting in a 38% decrease in the overall average job execution time.• Spearheaded cost reduction and storage optimization for AWS Redshift and Snowflake, using Python and Pandas for a 25% efficiency boost and significant cost savings across 9 regions.

Worldwide, Oo

Responsibilities:• Formulated and optimized extraction patterns using regular expressions to retrieve the relevant information, resulting in the determination of the specifics of each Loan Notice Document, increasing overall deliverability by 5%.• In collaboration with the Quality Assurance (QA) team, developed a service that automatically performs a set of tests for addressing issues in Azure DevOps using the Agile methodology, resulting in a three-fold increase in bug resolution speed.• Overhauled the OCR back-end which enhanced the overall extraction efficiency from 89% to 95%.Achievements:1) Recipient of Deloitte's Machine Learning Guild Program.2) Winner - New Hire Orientation

Worked in Back-end part of application development. Understood about the startup ecosystem. Technologies used: Python, Django, REST framework, Postman

Bangalore, Karnataka, In

Deep dived into Machine Learning algorithms and implemented the following:1) Sentiment analysis on Customer reviews(industry agnostic) to classify them as positive or negative. If a negative review is found, implemented a mechanism by which Business owners can improve their customer satisfaction. This task mostly involved Natural Language Processing.2) Implemented a Recommender system on amazon apparels. Build the system using different machine learning algorithms including K-NN, SVM and Neural Networks.

Master'S Degree, Computer Science

Bachelor'S Degree, Information Technology