# Kausshik Manojkumar

> Software engineer at RTX working on safety-critical flight management systems. Builds retrieval systems, native macOS apps, and ML pipelines. Published at ACM TKDD 2025.

## About

Kausshik Manojkumar is a software engineer at RTX (Collins Aerospace) in Cedar Rapids, Iowa. He is one of three engineers selected to define migration strategy for a 2-million-line Ada-to-C++ modernization of flight management software. He graduated from Iowa State University in May 2025 with a Bachelor of Science in Computer Science and a minor in Mathematics with Honors (GPA 3.97).

Before RTX, he was a machine learning research assistant under Dr. Yang Li at Iowa State University, where he contributed experimental results to an ACM TKDD 2025 publication on graph-based time-series forecasting with Graph Sequence Attention.

He is interested in problems at the intersection of systems programming, AI/ML, and native application development. He cares about software that is correct, fast, and maintainable.

## Contact

- Email: kausshikmanojkumar@gmail.com
- GitHub: https://github.com/KAUSSHIK
- LinkedIn: https://linkedin.com/in/kausshikm
- Website: https://kausshik.dev
- Resume: https://kausshik.dev/Kausshik_Manojkumar_Resume.pdf

## Work Experience

### Software Engineer I — RTX (Collins Aerospace) (July 2025 — Present)
Cedar Rapids, IA. Safety-critical flight management systems.
- Selected as 1 of 3 engineers to define migration strategy for a 2M+ LOC Ada-to-C++ modernization of flight management systems.
- Performed dependency analysis across 12+ behaviors and authored technical documentation defining sequencing, parallelization, and phased migration — now used as reference by implementation teams.
- Reimplemented core legacy flight-plan functionality in modern C++, reducing code complexity and improving maintainability.
- Co-developed an internal PyTest framework with new preconditions and verification paths, reducing manual testing effort by ~40%.

### Machine Learning Research Assistant — Iowa State University (May 2024 — April 2025)
Under Dr. Yang Li.
- Drove experimental expansion for a graph-based time-series forecasting system across NYC Cab, energy load (ECL), and road traffic datasets.
- Built end-to-end experiment pipelines: preprocessing, graph inference via Gaussian Markov Random Fields, multi-horizon forecasting, and evaluation.
- Contributed benchmark results to an ACM TKDD 2025 publication demonstrating state-of-the-art accuracy.
- Benchmarked Graph Sequence Attention vs. standard attention mechanisms.

## Projects

### QuillReader — Native macOS PDF reader (In Development)
- URL: https://kausshik.dev/project/quillreader
- GitHub: https://github.com/KAUSSHIK/QuillReader
- Tech: Swift, SwiftUI, PDFKit, CryptoKit, Keychain, SSE streaming
- Privacy-first PDF reader with AI-powered annotations. Users control what data leaves their machine via a privacy dial (none / local / metadata / full-text). Non-destructive annotations persisted as sidecar JSON files keyed by SHA-256 document fingerprint. BYOAI provider protocol — no vendor lock-in.

### UniLibrary RAG System — Production retrieval-augmented generation system (Shipped)
- URL: https://kausshik.dev/project/unilibrary-rag
- GitHub: https://github.com/KAUSSHIK/unilibrary-rag
- Tech: Python, LangChain, Qdrant, n8n, GCP Cloud Run, Next.js, React, TypeScript
- Semantic search engine over 1,000+ academic PDFs for a university research archive. Two-stage pipeline: hybrid retrieval (dense vectors + metadata filters) followed by an LLM validation pass that catches hallucinated citations before responses reach users. Deployed to 3 departments. 95%+ citation accuracy, sub-2s latency.

### Time-Series Forecasting with Transformers (Research)
- URL: https://kausshik.dev/project/stock-transformer
- GitHub: https://github.com/KAUSSHIK/StockTransformer
- Tech: Python, PyTorch, NumPy, Pandas, Matplotlib, scikit-learn
- Custom Transformer architecture adapted for financial time-series. Sliding-window training to prevent information leakage. Rigorous baseline comparison against ARIMA and LSTM. Transformer outperformed ARIMA by 18% and LSTM by 7% on multi-step horizons (MSE). Identified failure mode: Transformer underperforms LSTM during high-volatility regime changes.

## Skills

- Languages: C++, C, Python, Swift, Java, TypeScript, SQL, Bash
- ML / AI: RAG pipelines, Transformers, time-series modeling, embeddings, fine-tuning, model evaluation
- Backend & Systems: Spring Boot, FastAPI, Flask, REST APIs, WebSockets, Linux, large codebases
- Data & Infrastructure: MySQL, DynamoDB, Qdrant, AWS (EC2, S3, RDS), GCP, Docker, Git
- Frameworks: SwiftUI, PDFKit, React, Next.js, LangChain, PyTorch, n8n

## Education

- Iowa State University — B.S. Computer Science, Minor in Mathematics with Honors. GPA 3.97. May 2025.
- Stanford University Online — Machine Learning Specialization. January 2024.

## Publications

- ACM TKDD 2025 — Graph-based time-series forecasting with Graph Sequence Attention (acknowledged contributor).