Amritansh Mishra

Machine Learning Engineer

San Jose, California, United States2 yrs 8 mos experience

Most Likely To Switch

Key Highlights

Expert in Large Language Models and Machine Learning.
Proven track record in Fintech solutions and backend development.
Strong academic background with a 3.98 GPA in MS CS.

Stackforce AI infers this person is a Backend-focused Machine Learning Engineer with expertise in Fintech and SaaS solutions.

Contact

Skills

Core Skills

Large Language Models (llm)Python BackendNatural Language Processing (nlp)Python (programming Language)KubernetesMachine Learning PrivacyMicroservicesAutomationJava

Other Skills

TritonPyTorchLangChainVectorDBDPOMLflowAmazon Web Services (AWS)DockerGPUNumPyOptimizationData PrivacyMariaDBDropwizardAirflow

About

As a Software Engineer and a MS CS student at Umass Amherst, I am passionate about solving real-world problems using cutting-edge technologies. I have a strong background in Backend Development, Neural Networks, Machine Learning, and NLP. Previously, I worked at PhonePe, India's leading digital payments platform, where I collaborated with various teams to deliver innovative and user-friendly solutions. I successfully implemented Login With Phonepe, a unified login feature integrated into various apps, enhanced user experience by expanding login options, led efforts in devising a number churn solution, and spearheaded the development of a comprehensive user dashboard. I also contributed to codebase improvements, database migration and optimization, and workflow automation using Airflow. I am eager to learn from and work with diverse and talented professionals.

Experience

2 yrs 8 mos

Total Experience

10 mos

Average Tenure

11 mos

Current Experience

Capital one

Senior Associate(ML Engineer/Applied Scientist), LLM Core and Agentic AI

Jun 2025 – Present · 11 mos · New York, NY · On-site

Worked on building a travel agentic AI chat conceirge system, which could help in booking flights, you could provide complex utterances, and the agent would be able to understand intent, plan an action plan and fetch relevant flight searches with user preferencs.
Worked on Evaluating performance of GPTOSS 20B and 120B on different reasoning modes, for conversations datasets. Benchmarked them by synthetically generating a conversational dataset.
Worked on training a router with reinforcement learning to route calls to best models based on chat history. This helped in saving costs to the company.

Large Language Models (LLM)Python BackendTritonPyTorchLangChainVectorDB

Balbix

Machine Learning Engineer

Jun 2024 – Aug 2024 · 2 mos · San Jose, California, United States · On-site

Built a pipeline to load a saved a model convert it to tensorrt llm engine and deploy the model to triton inference server.
Currently working on adding support for MultiLora deployments using TensorRT-LLM and triton inference server. This is expected to increase the inference of existing models.
Moved the models loading directly on the container to triton server using python backend, this reduced the CPU usage by over 25 percent and reduced costs by 3x.

Python (Programming Language)KubernetesMLflowTritonAmazon Web Services (AWS)Docker+2

University of massachusetts amherst

Graduate Student Researcher

May 2024 – Sep 2025 · 1 yr 4 mos · Amherst, Massachusetts, United States · On-site

Working with the IESL lab and Collaboration with IBM in the Project SAIL.Currently working on ways to improve LLM capabilities on reasoning tasks by looking at step level dpo loss. Aim is to imorove reasoning capabilities on LLMs on datasets like gsm8k.

Large Language Models (LLM)Natural Language Processing (NLP)DPOPython (Programming Language)

Google deepmind

Graduate Student Researcher

Jan 2024 – Jun 2024 · 5 mos · United States · Remote

Currently working as part of the course 696DS in the project 'Measuring the privacy properties of Zeroth Order Optimization'.
As part of the project implementing the zero order optimization algorithms in Image Classification and NLP tasks and figuring out how does that affect privacy of the model.
This is an industrial mentorship research project in collaboration with the Information Extraction and Synthesis Laboratory (IESL) at UMass Amherst.
Industry Mentor(s): Tian Li, Manzil Zaheer
PhD Mentor(s): Dzung Pham, Wenlong Zhao

Python (Programming Language)NumPyOptimizationPyTorchLarge Language Models (LLM)Machine Learning Privacy+1

Phonepe

3 roles

Software Engineer

Jun 2022 – Oct 2023 · 1 yr 4 mos · Bengaluru, Karnataka, India

Collaborated with the Users team to successfully implement Login With Phonepe, a unified login feature integrated into various apps, including Pincode. Streamlined the login process by creating a centralized account service for all sister apps.
Enhanced user experience by expanding login options to include email addresses as additional identifiers alongside phone numbers.
Led efforts in devising a Number Churn Solution, developing an automated workflow using Airflow to effectively manage and mitigate the challenges associated with number churn.
Spearheaded the development of a comprehensive User Dashboard, empowering Oncall teams to efficiently handle daily Jira tickets. By leveraging the dashboard for user data retrieval instead of making repetitive API calls, significant improvements in workflow efficiency were achieved.

Python (Programming Language)JavaMariaDBDropwizardAirflowMicroservices+8

Software Engineer Intern

Jan 2022 – Jun 2022 · 5 mos · Bengaluru, Karnataka, India

Collaborated closely with both the Users and P2P teams to ensure smooth coordination and effective communication throughout project lifecycles.
Developed and implemented an automated solution for handling number churn cases, aligning with the latest guidelines released by TRAI (Telecom Regulatory Authority of India).
Contributed to codebase improvements, including database migration and optimization, resulting in enhanced performance and increased test coverage.
Actively participated in the design and development of multiple Airflow jobs, leveraging the power of workflow automation to streamline processes and improve efficiency.

JavaDropwizardAutomation

Software Engineer Intern

May 2021 – Jul 2021 · 2 mos · Bangalore Urban, Karnataka, India

Worked with the Users team to build a Service pendency which was used to fetch the number of pending states of any System.The Service used to track and untrack callbacks and fetch the pendency Count.
Also made a data selector which selects states which have a less pendency as compared to others.
Worked on the core backend using dropwizard and aerospike as the database.Also wrote tests using junit5.

JavaDropwizardAerospike