Looking for a job

Neel Mishra

@neelmishra

Applied Scientist II at Microsoft building explainable ML risk scoring for 34 million applications.

At Microsoft, I productionized explainable ML risk scoring across 34 million applications. A graph-based approach I developed improved OAuth application risk-model recall from 82% to 91% at 90% precision.

At Myntra, I created the Maya GenAI chatbot, whose multi-agent design increased engaged MAU by 58.86%. I also implemented approximate-nearest-neighbor search algorithms adopted across the company’s vector search products.

At Tekion, I built a RAG pipeline for a vehicle marketing chatbot and fine-tuned LLMs for email generation. My research includes published work at IJCNN and WCCI, and an accepted EMNLP 2026 paper.

Experience

Work history, roles, and key accomplishments

Current

Applied Scientist 2

Current

Microsoft

Sep 2025 - Present (1 year 1 month)

Productionized explainable ML risk scoring across 34 million applications, improving recall from 82% to 91% at 90% precision. Migrated a cloud ML experimentation platform and extended risk-scoring coverage to 74 million applications.

Python PySpark Azure Synapse LightGBM SHAP Isolation Forest Graph Based ML Distributed Feature Engineering Azure ML Pipelines

Data Scientist

Myntra

Sep 2024 - Jul 2025 (10 months)

Created and deployed the Maya GenAI chatbot, achieving a 58.86% increase in engaged MAU and a 31.74% boost in repeat usage. Implemented HNSW and ANNOY algorithms and optimized recommendation systems.

GenAI Multi Agent Systems HNSW Annoy CLIP Word2Vec PySpark Recommendation Systems Vector Search

Data Scientist

Tekion

Jul 2023 - Aug 2024 (1 year 1 month)

Built a RAG pipeline using LangChain and FAISS for a vehicle marketing chatbot. Fine-tuned LLMs with LoRA, achieving 21% higher similarity and 35% improved relevance, and reduced inference time by 97%.

RAG Lang Chain Faiss LLM As A Judge LoRAs VLLM GPTQ Text Summarization Hypothesis Testing

Research Assistant

IIIT Hyderabad

Aug 2020 - Jun 2023 (2 years 10 months)

Formulated a novel deep learning solver achieving state-of-the-art 94.75% accuracy on models like ResNet 50, DLA, and VGG. Developed a minimax GAN optimizer with Microsoft India.

Deep Learning Optimizer Design Gan ResNet VGG DLA Minimax Optimization