Aakash Vashishtha

Machine Learning Engineer

Technologies: Python, PyTorch, HuggingFace, Gymnasium, SLURM

• Implemented ProPS framework using GPT-OSS 20B (20 billion parameters) on ASU Sol supercomputer, leveraging 80GB A100 GPUs for LLM inference during policy optimization
• Orchestrated 200+ LLM calls across 4,000 episodes using SLURM job scheduling and managed long-running experiments via tmux with accelerate launch for GPU allocation
• Evaluated across 15 continuous/discrete control tasks (MuJoCo, Atari, Classic Control) comparing linear policies optimized by LLM reasoning vs neural network-based baselines
• Engineered Jinja2 prompt templates for structured LLM communication, implementing regex-based parsing to extract policy parameters from natural language responses
• Produced technical documentation and implementation enabling reproducibility for the ProPS framework (contributed to NeurIPS-accepted research; ProPS outperforms PPO/TRPO on majority of tasks)

Software Development Intern

Technologies: Django, Celery, RabbitMQ, Neo4j, MongoDB, Flutter iOS, Next.js, TypeScript, REST APIs, Jenkins, ArgoCD, Datadog

• Built production-grade symptom APIs powering real-time doctor-patient workflows and automated end-to-end ingestion pipeline for Athena EMR health history forms using event-driven architecture with Celery and RabbitMQ
• Engineered data transformation system mapping Athena form definitions to internal symptom templates using Neo4j DataKeys with full audit trail, versioning, and admin UI for review/approval workflows
• Refactored monorepo by extracting shared ServerAction and ServerFormAction classes into @advinow/server-actions package, eliminating code duplication and standardizing authentication package exports across Doctor and Conductor apps
• Developed dynamic form rendering system in Flutter patient app enabling runtime UI generation from JSON configurations, supporting clinic-specific flows without code deployments
• Achieved 2x faster patient intake flow by implementing REST API integrations, form validation, and iOS-specific features via MethodChannel for cross-platform consistency

Software Development Analyst

Technologies: AWS (AppSync, ECS), GoLang, ReactJS, Docker, Python (LangChain), JSX, RabbitMQ, Jira, PostgreSQL

• Spearheaded migration to React and implemented ADA compliance, enhancing overall accessibility and user experience
• Developed ProxyAI using GPT-4 API, automating 60% of data entry tasks and leading to a 15% increase in user engagement
• Engineered APIs and front-end solutions for SSO data management, improving admin workflow efficiency by 40%
• Wrote comprehensive GUI testing scripts for the activities section, validating component flow and ensuring platform reliability
• Implemented AWS AppSync monitoring, reducing subscription incidents by 50% and optimizing API usage
• Designed and developed APIs for email logs, enabling admins to track bugs reported by customers more effectively

Capstone Project

Technologies: Python (Flask, LangChain), FAISS, RAG, ReactJS, Docker, MySQL

• Developed a Flask-LangChain-based system to fetch and analyze Reddit/X posts for real-time social media insights
• Implemented a RAG pipeline using FAISS to compare fetched posts with user-stored products, enabling intelligent content matching
• Integrated ChatGPT and DeepSeek to generate human-like responses for automated engagement
• Built an automated system to post AI-generated responses back on Reddit/X, streamlining social media management
• Developed a ReactJS frontend for seamless user interactions and managed MySQL in a Dockerized setup

Software Developer II

Technologies: AWS (Step Function, CloudWatch, Athena, Glue, Lambda), Python (Boto3, PySpark), PostgreSQL, JFrog, Java (Spring Boot), RabbitMQ, GCP

E-commerce Platform Development:

• Architected and developed cart and orders microservices using Java Spring Boot and Netflix Eureka in a microservices architecture
• Deployed microservices and databases on GCP, ensuring scalable and reliable infrastructure

Ingest and Publishing for Organization Authority:

• Modified Ingest Step Function to increase max concurrency for ingestion Glue jobs, significantly reducing execution time and improving efficiency
• Performed data validation and preprocessing to standardize data ingested in tables, ensuring data quality and consistency
• Created Glue jobs to generate reports (NAICS/country/SIC/partition) of organizations, storing results in S3 for customer-facing APIs with SNS notifications
• Increased test coverage from 7% to 80% for the entire GCS publishing repository and identified critical bugs causing publishing Glue job failures
• Spearheaded implementation of Spark executor logs for all Glue jobs, addressing a long-standing issue adopted across Org Authority Team repositories
• Engineered organization ID mappings in PostgreSQL via PySpark, optimizing performance and transitioning to a shared library for streamlined use
• Led a 4-member team in integrating Dragos on an AWS Outpost server with Greengrass, developing an end-to-end data pipeline for AWS QuickSight reports, utilized as a POC for clients like Shell and Koch

Platform Engineer

Technologies: Terraform, AWS (EC2, EventBridge, EMR, Athena, Glue, Lambda), Python (Boto3, PySpark), HiveQL, Kafka

Infrastructure as Code Migration:

• Migrated infrastructure to IaC using Terraform & Chef, enabling features such as spinning up Redshift, EMR & Kubernetes clusters
• Configured applications to run on Chef bootstrapped servers with AWS auto-scaling and load balancing (ALB & ELB) support

Next-Gen Data Lake Architecture (POC - Adopted Org-Wide):

• Ran Glue crawler and EMR jobs (PySpark and HiveQL queries) to benchmark efficiency on 125 GB data, documenting best practices
• Executed Glue jobs & Athena queries on schemas created by Glue crawler, establishing patterns for the new setup
• Increased system availability by automating fallback to backup clusters when EMR targets are unhealthy using AWS Lambda, Network-LB, and Boto3
• Created an email-based notification service on serverless architecture (Lambda, EventBridge, SNS) to notify teams when EMR clusters terminate and backup clusters activate

Software Engineering Intern

Technologies: Python (Django), MySQL, JavaScript, Bootstrap 5

• Developed Report Manager - a web application to automate ticket-approval records, eliminating manual Excel sheet management and enabling on-demand Excel/PDF/ZIP file generation
• Created Bitlocker - an inventory management app tracking company assets assigned to employees

Software Engineering Intern

Technologies: ElectronJS, Bootstrap 5, SQL/WebSQL

• Created front-end and Windows executable (ElectronJS) for POS (Point Of Sale) website
• Implemented state management using client-side cookies (WebSQL, JavaScript) with offline data storage and automatic database sync when users come online

Arizona State University, Tempe

GPA: 3.7/4.0

Data Visualization: Nutritional Impact of Fad Diets

Developed a Python-based machine learning model using Ridge Regression to analyze and predict calorie content in dishes, classify them into appropriate fad diets, and identify potential nutritional deficiencies. Implemented data-driven visualizations, including calories vs. carbs and FDA-recommended vs. actual daily vitamin intake, providing insights into diet trends and health impacts.

Unity Game: Midnight Maneuver

Developed a taxi simulation game where players navigate the city to pick up and drop off passengers while racing against time. Implemented reckless driving detection that triggers police chases, along with evasion mechanics, health management, and power-ups for strategic gameplay.

Spring Boot App: ScrumPlay

Developed ScrumPlay, a team-based Scrum simulator using Spring MVC and Java Swing. Mimicking a Microsoft Azure environment, users engage in sprints, scrum calls, and sprint planning meetings, allowing dynamic modifications to user stories. The application fosters collaborative project management, generating insightful burndown charts post each sprint.

Android App: VirtualDecor

Developed a Java application for buying showpieces using AR (Google AR-Core), allowing users to visualize items in their homes before purchasing. Implemented user profiles using Firestore for data persistence.

Data Mining: Machine Learning using Transformers

Implemented the paper "Attention is all you need" to translate between Hindi and English sentences. Used IIT Bombay English-Hindi Corpus as dataset and created word embeddings using Facebook's fast-text.

Java Game: Spykey

Developed an object-finding game using template matching (Java OpenCV) where users find given objects in crowded background photos as fast as possible to beat the leaderboard. Implemented MySQL database on AWS EC2 for leaderboard management.

C# App: Manipal Tourism Information System

Developed a tour management tool where users can view interesting destinations on maps, book hotels, leave reviews, book tour guides and tour packages. Built using Oracle SQL database and C# for front and back end.

iOS App: Allergy Card

Developed an iOS app (Swift) allowing first responders to access user allergy details and information (Firestore database) in emergencies via ID stored on NFC card (CoreNFC). Android users are directed to a webpage (AngularJS).

Web App: KnowDisaster

Led a team of 3 to develop a web application to predict flood occurrence, notify users & emergency contacts, and show nearest relief camp locations. Built using VueJS, MongoDB for user profiles, WebSQL for geo-location queries, HeatmapJS and Google Maps API for location heatmaps, and AccuWeather APIs for weather data.

Robotics: R.U.G.V.E.D

Member of AI subsystem working on simulations & lane detection for RUGVED Systems, a student team designing autonomous robots with cutting-edge artificial intelligence.

• Simulated IGVC ground and RUGVED autonomous vehicle using ROS design engine, Gazebo, ran simulations using actual vehicle measurements to test trajectory using 3D visualization tool RViz
• Wrote code for Lane Detection by filtering camera feed on RUGVED using Python OpenCV