SOFTWARE ENGINEERING · APPLIED AI
Advaith Malka
Software engineer building
AI systems and products.
Projects
Traffic Simulator
Real-time 3D multi-agent traffic simulator using the Improved Intelligent Driver Model and MOBIL lane-change logic to surface emergent behavior like phantom jams.
ViewDATA PIPELINE
eBay + Amazon → JSON → OpenAI → Database
Product Data Ingestion System
Fault-tolerant AWS pipeline that turns product pages into schema-validated metadata; benchmarked at about 3,813 events per second.
View
StructAI
Fine-tuned version of Meta LLaMA that can generate real-time visualizations of data structures.
View
Cop Classifier
Classifies cars as police vehicles or regular traffic. Can be used in traffic signals to detect cop cars.
View
MediCord
An app that transcribes and summarizes audio files quickly and efficiently for patient use after doctor visits.
View
JobRecover
A web app made for HackTJ 7.5 and designed to connect employers to the unemployed, during difficult times. Thanks to Sritan and Nithil for making this project possible.
View
A little about me.
Hi! I’m Advaith Malka, a Computer Science senior at Virginia Tech focused on machine learning and AI systems. I’ve built everything from fine-tuned LLMs to computer vision pipelines, and I’ve shipped production AI tools at Collins Aerospace (RTX). When I’m not coding, you can find me on the court playing basketball, pickleball, or tennis.
Where I’ve worked.
Cybersecurity and AI Engineering Intern
RTX
Cedar Rapids, IA
May 2025 – August 2026
- Built Neo4j knowledge graph and AI Agent pipeline to generate RMFs, reducing compliance writing time by 50%.
- Developed AI agent with RAG for context-aware internal-library code generation, speeding development by 70%.
- Built LLM-guided fuzzer that compared GRM/DUT outputs and found 20% more unique crashes than AFL++.
- Containerized 3 RAG services with Docker Compose, cutting setup ~80% and saving ~2 engineer-hours/deployment.
- Fine-tuned an LLM to emit uncertainty tokens, triggering RAG only below thresholds and cutting retrieval ~30%.
- Engineered a RAG evaluation pipeline using vLLM and ElasticSearch to benchmark LLMs on 30K+ queries.
- Outperformed AdaRAG, DRAGIN, and 4+ RAG methods on 5 datasets while cutting inference latency by 18%.
- Automated end-to-end data processing pipelines to standardize 6 major academic datasets for experimentation.
Undergraduate Research Assistant
EEG2TEXT
Blacksburg, VA
February 2025 – June 2025
- Improved EEG-to-text BLEU/ROUGE scores ~5% through multimodal pipeline optimization.
- Refactored and optimized 2000+ lines of code into modular components for improved readability and maintainability.
- Resolved 5+ data and pipeline issues by developing validation scripts and debugging multimodal EEG processing.
Skills
Languages
- Python
- Java
- C
- HTML
- CSS
- JavaScript
- SQL
Frameworks & Libraries
- PyTorch
- NumPy
- Pandas
- JakartaEE
- React
- Django
- Node.js
- MongoDB
- SASS
- Tailwind CSS