Yash Saraswat

AI Researcher

Noida, Uttar Pradesh, India3 yrs 7 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Led development of a high-traffic LLM-powered chatbot.
Achieved significant cost and latency efficiency improvements.
Expertise in multimodal AI systems and NLP.

Stackforce AI infers this person is a Machine Learning Engineer specializing in SaaS and AI-driven solutions.

Contact

Skills

Core Skills

Machine LearningGenerative AiNatural Language Processing (nlp)Software DevelopmentCloud Computing

Other Skills

GPT-3PythonGPT-4LLMsPrompt EngineeringFine-tuningBERTSelfInstruct frameworkJavaAWS LambdaFinetuning LLMsParameter Efficient FinetuningTransformersLarge Language Models (LLM)GPT

About

Machine Learning Engineer with experience building production-scale AI systems across Generative AI, LLMs, Multimodal RAG, NLP, and Computer Vision. Currently working at Adobe on multimodal understanding and retrieval for Acrobat AI Assistant and Spaces. Previously at IndiGo, where I led backend development of an LLM-powered multi-agent assistant handling 80K+ daily requests with significant gains in cost and latency efficiency. Ex-SDE at Amazon. M.Tech in CSE (ML specialization) from IIT Kanpur.

Experience

Adobe

Machine Learning Engineer 2

Apr 2025 – Present · 11 mos · Hybrid

Working on multimodal understanding and retrieval for Acrobat AI Assistant and Spaces

Machine LearningGenerative AI

Indigo (interglobe aviation ltd)

Data Scientist

Apr 2023 – Mar 2025 · 1 yr 11 mos · Gurugram, Haryana, India · On-site

1. 6Eskai: An LLM powered Autonomous Agent
◦ Led the backend development of a GPT-4 powered chatbot 6Eskai incorporating power of LLMs like GPT-3.5 and GPT-4 for flight related services serving 80K requests daily.
◦ Optimized prompts to reduce GPT cost by 70% and also reduced GPT computation time by 40% from baseline.
◦ Developed a versatile framework aimed at faster creation of autonomous agent workflows.
2. LLM Finetuning
◦ Developed a cost-efficient autonomous agent by fine-tuning Llama3-8B-Instruct and Phi-3-mini-Instruct models using the SelfInstruct framework, achieving upto 90% reduction in computational cost for handling simple queries and significantly improving query processing speed.
◦ Developed and implemented a fine-tuned BERT model for customer query classification, achieving a 20% improvement in accuracy compared to the zero shot inference on GPT-4.

GPT-3PythonNatural Language Processing (NLP)Generative AI

Amazon

Software Engineer

Jun 2022 – Mar 2023 · 9 mos

1. Workflow Automation and Monitoring
◦ Reduced manual monitoring time of deployment pipeline by 30% by making the pipeline full CI/CD.
◦ Worked on Optimization and workflow automation for various tasks resulting in speedup of upto 5X.
◦ Created design docs, SOPs, unit tests and integration tests for various tasks.
◦ Made the monitoring pipeline more robust by adding metrics and alarms for frontend and backend services enabling developers to work on critical issues on time without impacting customer experience.

JavaAWS LambdaSoftware DevelopmentCloud Computing