Aditya Varma

About Aditya Varma

AI/ML Engineer/ Data scientist with 5+ years of experience, specializing in building enterprise AI solutions that drive real business outcomes. Currently at Intuit, I architect LangGraph-based multi-agent systems processing daily interactions across TurboTax and QuickBooks platforms. My expertise spans the entire AI lifecycle—from designing stateful conversation management and RAG-enhanced solutions to deploying production systems achieving 99.9% uptime. Recent wins include preventing $15M in annual fraud losses through ensemble ML models at BlueSky Insurance, reducing telecom customer churn by 12% using predictive analytics, and automating 60% of customer support workflows through intelligent agent orchestration.I thrive at the intersection of innovation and practical application, transforming complex business challenges into scalable AI solutions. My technical toolkit includes LangGraph, LangChain, GPT-4, Claude, PyTorch, and cloud platforms (AWS/GCP), always ensuring regulatory compliance and explainable AI for enterprise deployments. Beyond building models, I\'m passionate about democratizing AI and mentoring teams to adopt cutting-edge GenAI practices—whether implementing multi-agent coordination patterns, establishing MLOps best practices, or translating technical complexity into business value. Based in Plano, TX, I\'m always eager to connect with fellow AI practitioners and discuss the future of generative AI in enterprise applications.

Experience

1. Generative Ai Engineer

   Intuit

   Aug 2024 — Present · Plano, TX, US

   Architecting Intuit Assist, an enterprise AI platform powered by GenOS serving 100M+ users across TurboTax, QuickBooks, and Credit Karma. Built autonomous financial agents using LangGraph with parallel execution and checkpointing - including Accounting Agent (95% transaction categorization accuracy), Finance Agent (85% accurate 30-day cash flow predictions using LSTM networks), and Payments Agent (helping businesses get paid 45% faster). Pioneered Reflexion agents with episodic memory and self-refinement loops enabling AI to learn from failures through verbal reinforcement, improving accuracy by 45%. Developed production RAG pipelines using hybrid search (BM25 + dense retrieval), cross-encoder reranking, and contextual compression across 10M+ documents achieving sub-200ms latency. Fine-tuned financial LLMs with LoRA/QLoRA on proprietary datasets, implemented multi-modal AI systems using Vision Transformers and OCR for invoice processing, and created semantic caching layers reducing LLM costs by $50K monthly.

Education

• REVA University

  Bachelor's, Electronics and Communication Engineering

• CHRISTIAN BROTHER UNIVERSITY

  Master's, Computer and Information Science