Software Engineer – AI/ML @ Amazon Ex–Thomson Reuters | Ex–Morgan Stanley Senior Machine Learning Engineer | Generative AI & LLM Systems
I build production-grade AI/ML systems—from data and models to infrastructure and real-world impact. My work spans multi-modal LLMs, agentic workflows, large-scale MLOps, and applied ML systems serving millions of users.
⸻
• Design and deploy end-to-end ML & LLM systems at scale
• Build multi-modal, agentic AI workflows for real business problems
• Optimize model performance, cost, and latency in production
• Lead MLOps pipelines for fast experimentation and reliable deployment
⸻
Amazon (Social Ads Team) | Aug 2024 – Present Seattle, WA • Built a full-stack ML system from scratch to production to evaluate influencer-generated ad content quality using Amazon Bedrock and multi-modal LLMs, improving content review speed by 40%. • Orchestrated an agentic workflow using the open-source Strands SDK, integrated with API Gateway and AWS Lambda, automatically scoring image, video, and metadata. • Scaled the system to process 2M+ products daily with high reliability and low latency. • Integrated TikTok Trends API to identify trending content and generate ads aligned with real-time trends, resulting in a 20% uplift in overall ad performance (OPS) versus standard catalog-based ads.
⸻
Thomson Reuters | Sep 2023 – Aug 2024 Remote • Pioneered a state-of-the-art Legal Language Model (LLM) by curating and processing 10TB+ of legal data (LegalBench, proprietary corpora), outperforming LLaMA and Mistral by 15% on legal NLP benchmarks. • Built a comprehensive AI platform for contract summarization, legal Q&A, and text generation, delivering context-aware legal insights. • Led the end-to-end MLOps lifecycle—data ingestion, training, evaluation, deployment—reducing training time by 40% and enabling experimentation across 100+ model variants. • Applied task classification, multi-task learning, and few-shot prompting, achieving GPT-4–level performance on domain-specific benchmarks while using 30% less compute on RTX 3090, V100, and H100 GPUs.
⸻
Morgan Stanley | Aug 2020 – Jul 2022 Bengaluru, India • Designed a responsive dashboard for asset managers to manage client profiles and export insights to PDF/PPT, driving a 50% business increase and $2M+ in revenue. • Built a hybrid Ionic application to manage emails, calls, and meetings—onboarding 30,000+ users with ~70% daily active usage. • Implemented Cypress-based test automation, reducing production bugs by 30%. • Developed systems to track and analyze client interaction data stored in NoSQL databases.
⸻
Research | NLP | Transformers • Built a hybrid rule-based + ML system to convert the point of view (POV) of voice assistant messages across multiple languages (Alexa, Siri, Google Assistant). • Trained sequence-to-sequence transformer models (T5) achieving a BLEU score of 65.9. • Awarded a State-of-the-Art (SOTA) badge on Papers With Code for best-reported results. • Published research and open-sourced the work, demonstrating strong applied NLP and research rigor.
Computer Vision | Transformers | PyTorch | Hugging Face | Gradio • Replicated the seminal paper “An Image is Worth 16×16 Words: Transformers for Image Recognition at Scale” from first principles to deeply understand the Vision Transformer (ViT) architecture. • Implemented ViT end-to-end from scratch, including: • Image patching & flattening • Learnable patch and positional embeddings • Transformer encoder blocks with Multi-Head Self-Attention (MSA) and MLP layers • Validated correctness by training on the Pizza–Steak–Sushi dataset, reproducing the paper’s core learning behavior.
• Built and deployed Food Vision Big, a real-world food image classifier supporting 101 food categories using the Food101 dataset (101K images).
• Leveraged EfficientNetV2-S with transfer learning for high accuracy and efficient inference.
• Trained locally on NVIDIA RTX 4060, demonstrating large-scale CV training on consumer hardware.
• Developed an interactive Gradio UI and deployed the application on Hugging Face Spaces for public access.
🔗 Live Demo: https://huggingface.co/spaces/prithviraj-maurya/food-vision-big
📓 Key Notebooks: • 08_pytorch_paper_replicating.ipynb — ViT paper replication • 09_pytorch_model_deployment.ipynb — Model training & deployment
⸻
Master of Science in Data Science Indiana University Bloomington | Aug 2022 – May 2024
⸻
Languages: Python, Java, SQL, R ML / AI: PyTorch, TensorFlow, Scikit-learn, Hugging Face, LLMs, Generative AI, NLP/NLU Data & MLOps: Apache Spark, MLflow, Docker, CUDA, Airflow Cloud: AWS (S3, Lambda, Bedrock, API Gateway), Azure Data: NumPy, Pandas, Statistical Modeling, A/B Testing, Data Mining Other: Elasticsearch, Hadoop, Data Visualization, Regression, Clustering, Anomaly Detection
⸻
• 🏅 AWS Certified Developer – Associate
• 🧪 Kaggle Expert
• 📚 PyTorch Docathon Contributor
• 📄 Published research on voice assistant NLP models
• 🚀 Built ML systems impacting millions of users in production
⸻
• 🔗 LinkedIn: https://www.linkedin.com/in/prithviraj-maurya
• 📧 Email: pmaurya196@gmail.com
• 🧠 Kaggle: https://www.kaggle.com/prithviraj7387
• 🧑💻 GitHub: You’re already here 🙂
⸻
⭐️ I enjoy turning complex ML problems into scalable, elegant systems—always excited to build impactful AI.
⸻