Experience
2025 — 2026
2025 — 2026
Seattle, WA
• Designed and trained large-scale Graph Neural Networks based causal discovery models, building a production-ready causal preprocessing and graph-generation pipeline using causal-learn to uncover directed dependencies, latent structure, and actionable intervention paths.
• Built an end-to-end neuro-symbolic pipeline (PyTorch + PyReason) for ticket-risk analysis: model training, graph construction, ML/heuristic fact emission, rule-based inference, and explainable traces/alerts.
• Built a high-performance knowledge-distillation pipeline that trains ensemble of lightweight sequence models (BiLSTM/Transformer) to replicate PyReason’s temporal neuro-symbolic reasoning, enabling millisecond-latency inference without running the reasoning engine.
2024 — 2025
2024 — 2025
Manhattan, New York, United States
• Built a Snowpark/Snowflake dropout-risk model processing 10M+ enrollment records; engineered 100+ features with PCA and autoencoder reduction; trained an ensemble (XGBoost, logistic regression, deep learning) with Bayesian hyperparameter optimization and SHAP interpretability—boosting accuracy 20% over baseline and increasing student retention 15% via insights on engagement, attendance, and performance.
• Built and shipped a domain-specific RAG engine for NYU courseware: LangChain + FAISS (sentence-BERT) retrieval feeding LoRA-tuned Hugging Face Transformers. Deployed an AI course Q&A micro-service in Docker on Kubernetes,with Airflow + MLflow handling versioned training, drift retriggers, and live metrics—delivering a 25 % accuracy lift at almost 300 ms p99 latency.
• Built a React + TypeScript admin UI and Node.js proxy that calls Oracle Analytics Publisher via REST, validates parameters (Zod), parses CSV and presents enrollment capacity, enrollment, and waitlist KPIs with role-based access and catalog item deep links in OAC, thereby reducing enrollment-related help-desk tickets by 25%
2022 — 2023
2022 — 2023
India
• Assisted in rollout of an AI-powered Advanced Authorization platform that scores ISO-8583 transactions (0-99) in sub-ms at 65 K tx/s across active-active data centers. Deployed an ensemble (bi-GRU, GBT, elastic-net rules) distilled to ONNX and retrained monthly on 2 PB, backed by a 500-feature real-time store and a Kubernetes + Istio mesh—cutting fraud loss > 30 bps, lifting approvals 6 bps, and trimming false positives 60 %. Instituted SHAP explainability, drift/ AUROC back-tests, blue-green canaries, and secured PCI-DSS & GDPR compliance for TensorFlow/ONNX streaming pipelines.
• Accelerated the back-end development of Barclays’ in-house reconciliation software - Match Monitor 4 (MM4), responsible for cash and trade reconciliations, including generating and tracking breaks post cash or trade transactions
• Optimized SQL scripts responsible for inbound and outbound data flow between IntelliMatch and external systems like Delphi. Thereby, significantly reducing daily manual efforts of L1 teams by at least 20%. Technology leveraged include C#, .NET framework, SQL Server, T-SQL, TeamCity, Jenkins, Jira, SDLC using Scrum Agile methodologies
• Utilized tools such as TeamCity and Jenkins for continuous integration and deployment (CI/CD) pipelines, improving software build and release cycles.
2021 — 2021
2021 — 2021
India
Full stack developer - SSO products built on ASP.NET frameworks
• Engineered Single Sign-On products built on .NET Core and ASP.NET frameworks (MVC and Web Forms) by implementing SSO protocols like SAML, OAuth and OpenID Connect. Delivered continuous product feature additions, plugin version updates, usability enhancements, bug fixes, Nuget package creation & releases, user requested product customizations, seamless production deployment, etc. Accelerated revenue from .NET products by 50% in just 6 months.
• Single-handedly developed from scratch and released a SAML SSO plugin for e-commerce platform – nopCommerce. Was offered product ownership for the same. SAML Single Sign-On (SSO) - miniOrange - nopCommerce
2019 — 2021
2019 — 2021
Pune Area, India
• Advanced python scripts to incorporate features like performance monitoring, share provisioning & log analysis, as part of the distributed storage cluster functionality automation. Thereby, reducing system administration tasks by nearly 30%.
• Enhanced storage deployment workflows and ensuring configuration consistency using IaC: Ansible and Terraform
• Integrated CI/CD pipelines using Jenkins and GitLab to automate deployment of storage infrastructure changes, configuration updates, and monitoring solutions, reducing deployment time and ensuring consistency.
Education
NYU Courant Institute School of Mathematics, Computing, and Data Science
Master of Science - MS
Savitribai Phule Pune University