AI Engineer at Forethought

Developed LangChain multi-agent workflows with prompt engineering and chain-of-thought reasoning, plus API integrations (payment gateways, CRM), to automate real-time support actions, cutting ticket closure time by 60% and manual work by 40%. Created a self-updating ML system using vector embeddings to detect knowledge base drift and auto-merge resolved insights, improving accuracy by 25% (human-evaluated precision) and cutting new agent onboarding time. Replaced legacy RAG/evaluation frameworks with AIMon monitoring, drift detection, and A/B testing, while standardizing model serving patterns – improving LLM output accuracy on code-inclusive queries to 95%. Architected a low-code platform with Docker containerization and pluggable data source adapters, enabling teams to deploy AI agents across diverse SaaS products in under 2 weeks, increasing customer satisfaction by 30%. Resolved bottlenecks in real-time issue investigations by implementing semantic search using a vector database

Applied AI Engineer at Forethought

Tackled delays in resolving technical support tickets due to manual knowledge retrieval by developing an agentic AI assistant using retrieval-augmented generation (RAG) frameworks and integrations with tools like Slack, Jira, and Confluence for automated information gathering and solution generation, resulting in 50% faster ticket resolution time, 70% autonomous inquiry handling, and a 4.8/5 CSAT score in B2B SaaS environments. Addressed high-volume support queries requiring real-time actions by building multi-agent workflows with LangChain and API integrations for payment gateways and CRMs, leading to 60% streamlined end-to-end ticket closure and 40% reduction in manual intervention. Implemented a self-updating ML system to keep knowledge bases accurate, increasing KB accuracy by 25% and improving onboarding. Replaced legacy RAG and evaluation frameworks with AIMon-based monitoring and drift detection to boost LLM response accuracy for code-inclusive support. Designed a low-code platf

Machine Learning Engineer at Tally

Developed AI predictive analytics models (Scikit-learn, TensorFlow, time-series) on transaction data, achieving 85% forecast accuracy and 30% fewer errors. Engineered an intelligent document processing (IDP) pipeline using OpenCV OCR + fine-tuned BERT with REST APIs to automate invoice scanning, achieving 90% automation at 98% accuracy and saving 200+ manual hours monthly. Orchestrated real-time streaming pipelines using Kafka/Kinesis and Spark Structured Streaming to process 500k+ events/hour for feature engineering (including sentiment analysis) in recommendation systems. Implemented ML-powered QA bug detection models on AWS SageMaker, accelerating testing cycles by 35% and enabling earlier issue prediction in releases. Deployed anomaly detection ML systems that flagged data entry discrepancies using historical patterns, reducing compliance violations by 28% and improving data accuracy.

Software Engineer – AI/ML at 2 Tally

Addressed manual forecasting and data-entry bottlenecks by developing AI-powered predictive analytics models (scikit-learn, TensorFlow) achieving ~85% accuracy and 30% fewer forecasting errors. Automated 90% of image-based accounting entries using OCR with OpenCV and fine-tuned BERT for data extraction, delivering 98% accuracy and saving ~200 manual hours monthly. Built NLP-driven word-trend dashboards to analyze user feedback, delivering actionable insights and reducing analysis time by ~40%. Streamlined testing with AI-assisted tooling in SageMaker, cutting testing cycles by 35% and enabling early issue prediction. Implemented ML-based discrepancy detection to reduce compliance violations by ~28% and improve data integrity in end-to-end AI automation for accounting.

Software Engineer – AI/ML at Tally

Built AI-powered predictive analytics models using Scikit-learn, TensorFlow, and time-series techniques on transaction data, achieving 85% prediction accuracy and 30% fewer forecasting errors. Automated 90% of image-based accounting entries with OCR (OpenCV) and a fine-tuned BERT for data extraction, integrated via REST APIs, saving over 200 manual hours monthly with 98% accuracy. Developed an NLP-driven word-trend dashboard to analyze user reviews, delivering insights that boosted decision-making and shortened analysis time by 40%. Implemented AI tools for efficient bug detection and performance optimization using AWS SageMaker, accelerating testing cycles by 35% and enabling early issue prediction in AI-driven software release processes. Mitigated manual compliance and error highlighting in real-time accounting entries by designing ML systems to flag discrepancies, reducing violations by 28% and improving data accuracy.

Software Engineer – AI /ML at Tally

Developed AI-powered predictive analytics models using Scikit-learn, TensorFlow, and time-series methods to improve forecasting accuracy (85%) and reduce errors by 30%. Built an OCR pipeline with OpenCV and fine-tuned BERT for data extraction to automate 90% of image-based accounting entries with 98% accuracy, saving over 200 manual hours per month. Created an NLP-driven word-trend dashboard from user reviews to provide actionable insights, boosting decision-making speed by 40%. Implemented AI-driven bug detection and performance optimization in testing using AWS SageMaker, accelerating test cycles by 35% and enabling early issue prediction. Implemented anomaly detection for real-time accounting data entry to reduce violations by 28%.

Software Development Engineer Intern – AI/ML at Amazon

Addressed feature drift in production ML models by integrating basic validation scripts into CI/CD pipelines using MLflow and drift detection in an enterprise AWS environment, improving model reliability by 15%. Prototyped collaborative filtering models via matrix factorization in AWS SageMaker and offline evaluations to gauge engagement lift (CTR simulations) of ~8%. Optimized document processing workflows for cost efficiency with introductory active learning pipelines using Amazon Lex and SageMaker, reducing labeling costs by 40%. Helped mitigate inference latency in SaaS ML services by optimizing endpoints with AWS Inferentia and SageMaker autoscaling, cutting costs by 30%. Supported keyword extraction for content search, accelerating analyst processing by ~20% and gaining exposure to enterprise-scale software engineering in a large-scale setting.

Software Development Engineer Intern – AI/ ML at Amazon

Helped address feature drift in production ML models by developing basic validation scripts and integrating them into CI/CD pipelines with MLflow in an enterprise AWS environment, improving model reliability by 15%. Prototyped simple collaborative filtering models via matrix factorization in SageMaker and conducted offline evaluations to support engagement improvements. Assisted in optimizing document processing workflows for cost efficiency by building introductory active learning pipelines with Amazon Lex and SageMaker, reducing labeling costs by 40%. Gained experience in reducing inference latency by leveraging AWS Inferentia and SageMaker autoscaling, cutting costs by 30%. Participated in developing keyword extraction for content search using SageMaker, accelerating analyst processing by 20%.

Software Development Engineer – AI/ML at 2 Amazon

Improved reliability of production ML systems using data validation, drift detection, and model monitoring pipelines with Python, MLflow, and Great Expectations integrated into CI/CD. Architected scalable recommendation system pipelines using collaborative filtering and personalization logic, driving engagement lift in simulations. Designed data labeling and active learning workflows integrating a feature store and model registry and exporting curated datasets to BigQuery, reducing annotation costs. Enhanced real-time inference infrastructure via TensorFlow Serving, Kubernetes auto-scaling, and GPU optimization to reduce latency and compute costs. Built and maintained Spark/Airflow/Kafka pipelines to load processed results into a Snowflake warehouse to improve analyst efficiency.