Soham Rajesh Choulwar

Incoming Intern at Qualcomm

Senior at ASU

schoulwa@asu.edu

Experience

Software Engineering and Machine Learning Intern

Walnutech PBCSep. 2025 - Nov. 2025

Improved scholarship match ranking by 42% across 45,000+ scholarships by building a generative AI scholarship matching platform with MCP, FastAPI, LangGraph, MindsDB, and AWS ECS Fargate and S3
Increased relevance for complex student profiles by designing a two stage retrieval pipeline using MindsDB semantic search for candidate recall and GPT 5 mini reranking for precision
Reduced deployment lead time by 55% by implementing an end to end RAG workflow with GPT 4 for attribute extraction and CI/CD pipelines that automated 90% of evaluation runs with Langfuse and LangGraph

AI/ML and Prompt Engineering Intern

BayerJun. 2025 - Aug. 2025

Increased data retrieval efficiency by 40% for 1,500+ employees by shipping a Python server implementing the Model Context Protocol (MCP), integrating Databricks, PostgreSQL, AWS S3, and Terraform to provide secure access to 100+ data tables
Architected and automated internal text heavy workflows for 2,000 employees by designing and deploying three AI assistants using prompt engineering, spaCy, NLTK, and Hugging Face Transformers to extract insights from unstructured text

Data Engineering Intern

V2Stech SolutionsMay 2024 - Jul. 2024

Cut P95 query latency by about 35% in an NLP recommender engine built with Python, txtai, and SVD through vector index tuning, query caching, and profiling driven performance fixes
Reduced false positives in client recommendations by about 25% by fine tuning a Llama 2 model with LoRA on domain data and serving it with FastAPI

Researcher and Undergraduate Teaching Assistant

Arizona State University, CodeLabAug. 2025 - Present

Improved simulated agent collaboration success from 63% to 80% by building a Stag Hunt simulation using the OpenAI GPT API and agents SDK to model Bayesian belief updates across six difficulty levels and thirty tasks
Architected automated experimental framework with statistical analysis pipeline, orchestrating 240+ multi-agent trials across six threshold parameters and generating data visualizations to quantify coordination patterns

Projects

Multi Modal AI Desktop Assistant

Reduced response latency to under 200 ms across 100+ real-time interactions by implementing an event-driven assistant backend with WebSockets + REST APIs, plus a context retrieval layer that dynamically injected only the most relevant prior intent, tool state, and user preferencesAchieved 95% accurate intent/type detection across app control, screen analysis, and reminder workflows by building a full-stack routing engine with an Express.js API + Next.js frontend

Data Pipeline, CodeFlow Spark Challenge Winner

Automated CSV cleaning, wellness analysis, and recommendation generation for uploaded datasets by building a Data Wellness pipeline with a TypeScript, React plus Vite frontend and gAit LLM transformationsArchitectured a fully serverless, scalable data processing stack by logging metadata and analysis outputs to AWS S3 and deploying the backend using AWS Lambda, AWS Glue, S3, and Amazon RDS

Education

Arizona State University (4+1)

Bachelor of Science in Computer Science, Minor in Data Science

May 2027 -

Skills

Java

Python

C/C++

JavaScript

TypeScript

SQL (PostgreSQL)

HTML/CSS

GraphQL

React

Next.js

Node.js

Flask

FastAPI

Django

JUnit

Material-UI

DynamoDB

MongoDB

Git

GitHub

Docker

Amazon Web Services (AWS)

Azure

Google Cloud Platform

Postman

Jira

Terraform

Databricks

NumPy

scikit-learn

PyTorch

TensorFlow

Hugging Face

LangGraph

Model Context Protocol (MCP)

Large Language Models

Machine Learning

Natural Language Processing

Agentic AI

Data Pipelines

Connect

github.com/Soham-1827

sohamchoulwar.vercel.app

Experience

Software Engineering and Machine Learning InternSep. 2025 - Nov. 2025

Walnutech PBC

Improved scholarship match ranking by 42% across 45,000+ scholarships by building a generative AI scholarship matching platform with MCP, FastAPI, LangGraph, MindsDB, and AWS ECS Fargate and S3
Increased relevance for complex student profiles by designing a two stage retrieval pipeline using MindsDB semantic search for candidate recall and GPT 5 mini reranking for precision
Reduced deployment lead time by 55% by implementing an end to end RAG workflow with GPT 4 for attribute extraction and CI/CD pipelines that automated 90% of evaluation runs with Langfuse and LangGraph

AI/ML and Prompt Engineering InternJun. 2025 - Aug. 2025

Bayer

Increased data retrieval efficiency by 40% for 1,500+ employees by shipping a Python server implementing the Model Context Protocol (MCP), integrating Databricks, PostgreSQL, AWS S3, and Terraform to provide secure access to 100+ data tables
Architected and automated internal text heavy workflows for 2,000 employees by designing and deploying three AI assistants using prompt engineering, spaCy, NLTK, and Hugging Face Transformers to extract insights from unstructured text

Data Engineering InternMay 2024 - Jul. 2024

V2Stech Solutions

Cut P95 query latency by about 35% in an NLP recommender engine built with Python, txtai, and SVD through vector index tuning, query caching, and profiling driven performance fixes
Reduced false positives in client recommendations by about 25% by fine tuning a Llama 2 model with LoRA on domain data and serving it with FastAPI

Researcher and Undergraduate Teaching AssistantAug. 2025 - Present

Arizona State University, CodeLab

Improved simulated agent collaboration success from 63% to 80% by building a Stag Hunt simulation using the OpenAI GPT API and agents SDK to model Bayesian belief updates across six difficulty levels and thirty tasks
Architected automated experimental framework with statistical analysis pipeline, orchestrating 240+ multi-agent trials across six threshold parameters and generating data visualizations to quantify coordination patterns

Projects

Multi Modal AI Desktop Assistant

Reduced response latency to under 200 ms across 100+ real-time interactions by implementing an event-driven assistant backend with WebSockets + REST APIs, plus a context retrieval layer that dynamically injected only the most relevant prior intent, tool state, and user preferences
Achieved 95% accurate intent/type detection across app control, screen analysis, and reminder workflows by building a full-stack routing engine with an Express.js API + Next.js frontend

Data Pipeline, CodeFlow Spark Challenge Winner

Automated CSV cleaning, wellness analysis, and recommendation generation for uploaded datasets by building a Data Wellness pipeline with a TypeScript, React plus Vite frontend and gAit LLM transformations
Architectured a fully serverless, scalable data processing stack by logging metadata and analysis outputs to AWS S3 and deploying the backend using AWS Lambda, AWS Glue, S3, and Amazon RDS

Technical Skills

Skills: Java, Python, C/C++, Go, JavaScript, TypeScript, SQL (PostgreSQL), R, HTML/CSS, GraphQL, React, Next.js, Node.js, Flask, FastAPI, Django, JUnit, Material-UI, DynamoDB, MongoDB, Git, GitHub, Docker, Amazon Web Services (AWS), Azure, Google Cloud Platform, Postman, Jira, Terraform, Databricks, NumPy, scikit-learn, PyTorch, TensorFlow, Hugging Face, LangGraph, Model Context Protocol (MCP), Large Language Models, Machine Learning, Natural Language Processing