Looking for a job

Vaishnavi Saundankar

@vaishnavisaundankar

AI engineer at Georgia State University who fine-tuned Microsoft’s TrOCR, reducing Character Error Rate from 1.79 to 0.31.

United States

Message

At Georgia State University, I fine-tuned Microsoft’s TrOCR on handwritten Devanagari word images, reducing Character Error Rate from 1.79 to 0.31. I also built a Hugging Face pipeline to load and preprocess the handwriting data.

I’ve built RAG and agentic AI projects for insurance policy Q&A, clinical conversations, and CRM deal-risk recovery. Earlier at Infosys Ltd. and Accenture, I developed CRM flows, integrations, and validation logic.

Experience

Work history, roles, and key accomplishments

Current

Volunteer Research Associate

Current

Georgia State University

Feb 2026 - Present (9 months)

Fine-tuned Microsoft's TrOCR using LoRA on handwritten Devanagari word images, reducing Character Error Rate from 1.79 to 0.31 and Word Error Rate from 1.60 to 0.61. Built a Hugging Face pipeline for data preprocessing and model evaluation.

TROCR LoRAs Hugging Face Transformers Datasets PEFT CER WER Model Evaluation Python

Senior Associate Consultant

Infosys Ltd.

Feb 2022 - Jul 2023 (1 year 5 months)

Built a guided OmniScript purchasing flow for a telecom client with FlexCards and DataRaptors. Wrote eScript on Siebel Business Component events to validate data and configured Siebel CRM workflows for accident processing.

Siebel CRM OmniScripts FlexCards DataRaptors Siebel EScript Workflow Configuration Data Validation

Application Development Analyst

Accenture

Nov 2018 - Jan 2022 (3 years 2 months)

Developed a Siebel EAI integration pipeline parsing an 800K+ line XML product feed into Oracle Siebel CRM. Implemented eScript business services to validate a multi-level product catalog, cutting manual processing time by 40%.

EAI XML Oracle Siebel CRM Siebel EScript Product Catalog Data Integration Root Cause Analysis Regression Testing