Hello, I'm
Ashish Sangale
AI/ML Engineer + Software Engineer
Impact Highlights
30%
Faster training
Optimized distributed LLM pipelines on WSE for measurable throughput gains.
100K+
Transactions
Infrastructure powering monthly merchant volume at scale.
500K+
Active users
Performance wins across streaming products at production scale.
4+
Years experience
20+
Projects shipped
10+
Technologies
Resume
Experience
Machine Learning Integration & Quality Engineer Intern
Cerebras Systems, Inc.
- Optimized distributed LLM training pipelines in PyTorch on the Cerebras Wafer-Scale Engine (WSE), tuning batch sizing and runtime configurations to improve training throughput by 30% while maintaining convergence stability
- Built modular Python tooling to manage LLM checkpoints, prompt cache artifacts, and experiment metadata, improving reuse, reproducibility, and iteration speed in large scale training and inference workflows
- Implemented automated diagnostic logging to track training stability and inference metrics (latency, throughput, cache hit rates), enabling early detection of regressions
- Designed agent-driven validation and stress-testing pipelines to evaluate long-context behavior, cache correctness, and system-level failure modes under real-world and adversarial prompt distributions
Software Engineer II
Airpay Financial Technologies
- Built and scaled REST and GraphQL APIs using Python (FastAPI) and Node.js, backed by MySQL and Redis, supporting 100K+ monthly merchant transactions across Airpay Vyaapaar, Yako, and Wakala with 99.9% uptime
- Transitioned tightly coupled backend components to microservices with infrastructure-as-code (Terraform), enabling parallel team development and decreasing feature release lead time by 30%
- Orchestrated asynchronous transaction workflows using AWS SQS, Elastic Load Balancers, and database connection pooling, increasing peak traffic handling capacity by 35% and reducing request timeouts under high concurrency
- Strengthened production reliability by introducing automated test coverage, performance profiling, and structured error handling, decreasing Sev-2 incidents by 25% and improving mean time to resolution
Software Engineer
Jio Platforms Ltd.
- Engineered responsive user interfaces for JioSaavn and JioCinema using React and TypeScript, optimizing component rendering and bundle size to reduce page load time from 2.2s to 1.8s for 500K+ active users
- Enhanced Java and Spring Boot services managing user session data and content metadata, lowering API response latency from 320ms to 240ms during peak streaming hours
- Integrated automated test suites and pre-merge validation checks into CI pipelines, improving release stability and reducing post-deployment hotfixes by 20%
- Optimized distributed service performance using Datadog APM for request tracing and latency analysis, resolving bottlenecks and sustaining 99.9% uptime during major content launches
Education
Master of Science in Computer Software Engineering
Arizona State University
Bachelor of Engineering in Information Technology
University of Mumbai
Skills
Featured Project
EduSummarizer
EduSummarizer turns dense academic content into clear, structured summaries with a RAG pipeline and a fast, interactive UI. It combines retrieval, prompt orchestration, and evaluation-friendly outputs to keep summaries grounded in source material.
Contact
I'm always open to interesting projects and conversations. Whether you have a question, a proposal, or just want to say hi — my inbox is open.
Currently open to AI/ML and software engineering roles.