Hello, I'm

Karthik Prakash

Master of Science in Intelligent Information Systems at Carnegie Mellon University. Passionate about Machine Learning, Generative AI, and building scalable intelligent systems.

Resume
Karthik Prakash

Education

Carnegie Mellon University

Master of Science in Intelligent Information Systems

Pittsburgh, PA • Expected Dec 2027

Relevant Coursework: Natural Language Processing, Deep Learning, LLM Systems, AI Agents, Generative AI.

RV College of Engineering

Bachelor of Engineering in Computer Science

Bengaluru, India • Jun 2026

GPA: 9.73/10. Relevant Coursework: Data Structures & Algorithms, DBMS, Linear Algebra, Probability.

Experience

Machine Learning Intern

Draup

Jan 2026 - Jul 2026
  • Slashed Milvus vector RAM by 75% via INT8 Quantization and Matryoshka Learning, maintaining a strict 98% recall.
  • Deployed a potion-base-8M on-the-fly cross-encoder, halving storage overhead and improving semantic re-ranking.
  • Calibrated dynamic classifier thresholds recovering 8% false-negative signals at 83.1% precision across the retrieval pipeline.

Research Intern (EDR Evasion)

Technical University of Applied Sciences (THWS)

Jan 2026 - May 2026
  • Engineered an RL engine using Deep Q-Networks to model complex endpoint attacks as a Markov Decision Process.
  • Exposed blind spots in 5 enterprise EDRs (e.g. SentinelOne), attaining a 26.6% mean detection rate and 100% evasion rate.
  • Developed a scalable C++/Python adversarial platform integrating ML to automate continuous enterprise EDR red-teaming.

Software Engineering Intern

Microsoft

May 2025 - Jul 2025
  • Implemented a scalable multi-intent query processing framework for the Tenant Admin Copilot.
  • Validated and adopted Chain-of-Thought prompting via empirical benchmarking against Multi-stage Prompt Chaining.
  • Designed a metadata-driven API mapping engine using OpenAPI Swagger for seamless, real-time dependency resolution.

Featured Projects

Knowledge Graph-Enhanced RAG

Constructed a knowledge graph of 1,427 nodes via Neo4j and BERT embeddings for cyber threat intel. Implemented LLaMA 3.1 RAG achieving 87.2% hit rate.

Python LLaMA 3.1 Neo4j Pinecone

Intelligent Speech-Driven IR

Created a multimodal speech-to-text pipeline yielding 7.08% WER for multilingual queries. Integrated BERT and FAISS for scalable vector search.

PyTorch Hugging Face FAISS

ML Predictive Simulation Pipeline

Designed deep learning models predicting aerodynamic performance from 3D point cloud data. Replaced expensive CFD simulations with fast ML surrogates.

Deep Learning 3D Data RL

Publications

VYUHA: EDR Evasion via Deep RL-Guided Orchestration

ACM CCS 2026 (Submitted) • First Author

DL for Dementia Detection using Non-Invasive Data & Genetic Risk

IEEE CMSS 2025 • First Author

Pulse Point: Medical Data Intelligence for Human-Centric Interaction

IEEE ICOCT 2025 • First Author

Lightweight Machine Learning for Passive Acoustic Monitoring...

IEEE ICAISS 2026

Leadership

President

ACCELERATE RVCE

Nov 2024 - Dec 2025

Led the club's initiatives, organizing events and driving community engagement.

Member

ACCELERATE RVCE

May 2023 - Nov 2024

Active member contributing to various club projects and workshops.

Technical Skills

Languages & Frameworks

Python C/C++ SQL R Bash PyTorch TensorFlow

AI / ML / Data

LLMs / RAG Hugging Face LangChain Vector Databases OpenCV Pandas / NumPy Scikit-Learn

Cloud & MLOps

AWS / GCP Docker / Kubernetes Linux Git / CI/CD MLflow