Rohith Shinoj Kumar
Software Engineer | Exploring Scalable Systems & ML
I'm a final-year Computer Science and Engineering student at the National Institute of Technology Karnataka. I’m passionate about building scalable machine learning systems. With a strong foundation in Python, C++, and ML, I’ve worked on end-to-end projects across MLOps and full-stack development — turning ideas into deployable, impactful solutions. Always eager to learn, collaborate, and innovate. Let’s connect!
Work Experience
Centre for Development of Telematics (C-DOT)
Research Engineer (Incoming)
August 2025 — Present- Yet to join
Accenture Inc.
Software Development Engineering Intern
May 2024 — July 2024- Designed and deployed a scalable dynamic pricing pipeline, leveraging real customer analytics data and Facebook Prophet for time-series demand forecasting — improving forecast accuracy to 94%
- Implemented Bayesian A/B testing for price optimization,in leading to a 12% lift in customer conversion rates by continuously aligning prices with predicted demand.
- Developed a full-stack application integrating ReactJS frontend with Flask backend APIs, automating real-time pricing updates future scalability.
Outside the Stack (OTS)
Member of Project Team - Remote
Feb 2024 — May 2024- Optimized buffer utilization and reduced latency in a TLS proxy pair, enabling secure connections between clients and servers with deprecated SSL versions.
- Developed a high-performance logging system using memory-mapped JSON files, replacing the intial system logger.
- Cut read-write latency bynearly 20% and peak CPU usage by 8%, boosting proxy performance and reliability on new connections.
Featured Projects
ZAlpha - GenAI Powered Stock Portfolio Enhancer
- Built a portfolio analytics platform with advanced risk models (VaR, CVaR, MPT) for real-time multi-asset analysis and optimizations
- Integrated Ollama-hosted local LLMs and Finnhub API data for live and historical financial data, generating contextual strategy insights and AI-powered financial narratives.
- Built interactive dashboards using AI-powered tooltips, enabling real-time performance attribution, factor exposure analysis, and contextual insights from market data and portfolio behavior.
Python Ollama (Local LLMs) REST APIs Flask react.js FinTech
Reliable, Auditable Messaging Platform for Regulated Industries
- Designed and implemented a scalable, fault-tolerant messaging backbone for regulated industries (like healthcare, finance, etc) requiring guaranteed delivery, end-to-end audit trails, and zero data loss in high-availability environments.
- Used Kafka for durable event streaming and Redis as a low-latency replay + dead-letter queue system with structured audit logging to PostgreSQL.
- Built a dockerized deployment stack including Kafka, Redis, Postgres, and the app server for reproducible local and cloud environments.
Kafka Redis PostgreSQL Python Docker FastAPI
Transformer-Based Multimodal Pomegranate Disease Detection (ICAR-affiliated)
- This project is part of a funded joint initiative between NITK and the Indian Council of Agricultural Research (ICAR) focussed on deep-learning aided crop disease detection
- Developed a semi-supervised learning pipeline combining EfficientNet-CNN backbones with Vision Transformer (ViT), leveraging attention-based region focusing to achieve 95.7% classification accuracy despite a small non-standard dataset of ~500 field-collected leaf and fruit images across multiple disease categories.
- Worked in a team of 3 to deliver a fully optimized ONNX-quantized model (for cross-platform deployment) to ICAR for upcoming integration into their field-level diagnostic tool.
Semi-Supervised Learning ONNX PyTorch TorchVision ViT