Built AI-powered leadership enablement agents at TARIY using AutoGen, LangChain, LangGraph, and LangSmith to deliver personalized coaching, real-time roleplay simulations, and adaptive training experiences through multi-agent orchestration. Collaborated with cross-functional teams and business units to implement agents as Microsoft Teams plugins, enabling seamless integration into existing workflows.
Architected comprehensive ML pipelines and ETL workflows, designed predictive models using XGBoost and neural networks for customer analytics and fraud detection. Built scalable data processing infrastructure with Tableau, Snowflake, SQL, Python, achieving 15% improvement in model accuracy. Collaborated with different product teams to deliver end-to-end data solutions across multiple business units.
Optimized ETL workflows for the School Fuel database using an event-driven architecture with 5+ AWS services (Lambda, S3, Glue, Athena, IAM), integrated real-time alerts via SNS, and enhanced monitoring with CloudWatch—achieving up to 40% faster processing and 25% improved reliability.
Developed interactive visualization dashboards, handled missing data inconsistencies, automated workflows, collaborated with Data scientists to improve the model accuracy and maintain data integrity.
Built production-scale AI-powered e-learning platform processing 50+ educational videos with automated note generation using AWS Transcribe and fine-tuned GPT-3.5-turbo. Implemented vector-based Q&A system with FAISS and RAG architecture, achieving 90% accuracy in content extraction and 85% relevance in automated assessments with <150ms response time.
Built web scraping agent using Firecrawl, HTTP calls, and Agentic Tool Architecture to extract and answer queries from top 50 NASDAQ company reports. Integrated GPT-3.5-turbo (7B) via Azure OpenAI with LangGraph, FAISS, Postgres, and n8n; achieved <200ms latency and 24% gain in QA precision.
Built real-time stock market data pipeline processing 1M+ records daily using Apache Kafka, AWS services (S3, Athena, Glue, Lambda), and Python. Implemented streaming analytics with 99.9% uptime, automated ETL workflows, and interactive dashboards for comprehensive market analysis and predictive insights.
Developed a RoBERTa-based recommendation system analyzing 50K+ social posts for customer segmentation, achieving 92.7% accuracy and reducing latency from 250ms to 35ms using model distillation and TorchScript. Fine-tuned with LLM(RoBERTa) consisting of 100M+ tokens and optimized model efficiency by leveraging transfer learning and deploying a lightweight version on AWS EC2 with auto-scaling.
Built production-scale search engine using Flask, GPT-3.5-turbo, and vector embeddings processing 10K+ queries daily. Implemented semantic search with FAISS, real-time web scraping, and deployed on AWS Fargate (ECS) with auto-scaling, achieving <300ms response time and 94% user satisfaction.
Built production-scale recommendation system processing 10M+ movie records using Two-Tower architecture and SVD collaborative filtering. Integrated LLM-powered content generation achieving variability and personalized recommendations improvement with 92% recommendation accuracy using Amazon SageMaker and S3 storage for efficient deployment.
Developed interactive Tableau dashboard analyzing 2M+ flight records across all US airlines, processing real-time delay data with 99.5% accuracy. Implemented advanced analytics with Python, SQL, and Tableau Server, enabling stakeholders to identify delay patterns and operational inefficiencies, resulting in 15% improvement in decision-making speed.