ML Engineer · Building RAG pipelines, agentic AI & forecasting systems · Research @ IISc & NKUST · TEEP Scholar · Novartis Top 12
Building ML systems that ship — from power grid forecasting at IISc Bengaluru to production RAG pipelines serving enterprise clients in the Netherlands. My stack: XGBoost · LightGBM · Transformers · LangChain · LangGraph · CrewAI · FastAPI. I care about the full loop: feature engineering → model dev → low-latency inference → monitoring. Lately I've been designing multi-agent and HITL workflows for LLM-powered search — and co-authoring a paper currently under review at International Journal of Medical Informatics. A few things I'm proud of: → Top 12 National Finalist, Novartis Hackathon 2025 (out of 1000+ teams) → TEEP Scholar — Taiwan government research fellowship at NKUST → Research at IISc Bengaluru (Kotak-funded, highly competitive) Open to full-time ML Engineer or Applied Scientist roles. DM me or reach out at [email protected]
National Kaohsiung University of Science and Technology
1 yr · Kaohsiung City, Taiwan
• Selected as a TEEP Scholar — Taiwan government-funded international research fellowship awarded to top international candidates • Conducting research in physiological AI, recommendation systems, and multimodal machine learning at NKUST under Prof. Kang-Ming Chang • Applying transformer-based models, deep learning, and time-series regression techniques on HRV and EDA wearable sensor datasets for personalized health systems • Collaborating with a cross-cultural research team and presenting findings during weekly lab reviews and international research discussions • Co-authoring a PRISMA 2020 systematic review on AI-based biofeedback systems, currently under review at the International Journal of Medical Informatics
Indian Institute of Science (IISc)
2 yrs 7 mos · Bengaluru, Karnataka, India
• Accepted into highly competitive Kotak-funded research program at one of India's top research institutions • Built end-to-end ML pipeline for time-series forecasting on power grid data using XGBoost and LightGBM • Engineered features from smart-meter datasets; implemented preprocessing, lag features, and rolling-window transforms • Developed low-latency inference system for real-time energy demand prediction • Designed and evaluated multi-model experiments; documented results for publication
Accenture
9 mos · Bengaluru, Karnataka, India
The virtual internship at Accenture in Strategy Consulting provides a unique and immersive experience for participants to engage in real-world consulting scenarios from the comfort of their own workspace. This program is designed to mirror the challenges and responsibilities faced by strategy consultants at Accenture, offering interns a valuable glimpse into the dynamic world of strategic decision-making and problem-solving.
IIDE - The Digital School
3 mos · Bengaluru, Karnataka, India
Planet One
3 mos · Bengaluru, Karnataka, India
During my virtual internship at Planet One, I had the exciting opportunity to delve into the intricacies of SEO and Strategy Marketing, gaining invaluable hands-on experience in a dynamic and innovative environment. The internship provided a comprehensive overview of digital marketing practices, with a strong focus on search engine optimization and strategic planning.
NKUST — TEEP Government Scholarship
Feb 2025 – Nov 2025 · Remote, Taiwan
Architected LangGraph-compatible ranking and retrieval logic with Human-in-the-Loop gates; wrote specifications and acceptance criteria. Built multimodal time-series regression models on HRV/EDA wearable data and owned the model lifecycle through production evaluation and validation. Designed agent reliability metrics and iterative validation loops, tracking retrieval precision and recall across model versions. Conducted EDA on 100+ multi-sensor studies to surface clinically meaningful trends and guide model design.
Indian Institute of Science (IISc) AI & IoT Lab
Jan 2024 – Dec 2024 · Bengaluru
Trained XGBoost, LightGBM, and Random Forest models for time-series anomaly detection on power grid data and monitored data and concept drift. Built automated GitHub Actions CI/CD retraining pipelines with reproducibility and auditability. Fine-tuned and optimized Vision Transformers for document parsing through pruning and quantization for FastAPI deployment. Ran model comparison tests, shipped the winning variant to production, and built scalable HDF5 pipelines for GPU-scale training.
Product-Based SaaS Company
Jun 2024 – Sep 2024 · Remote, Netherlands
Architected and owned a production RAG pipeline using LangChain, HuggingFace embeddings, ChromaDB, and FAISS from specification through live deployment. Implemented ranking logic and confidence scoring with threshold-based filtering to mitigate hallucinations and reduce irrelevant retrievals. Achieved semantic similarity above 0.87 on multilingual Hindi and Kannada documents. Deployed Plotly and FastAPI dashboards translating ML outputs into business metrics, while independently handling architecture, latency optimization, and production error handling.
IBM Technologies
Jul 2024 – Aug 2024 · Remote
Deployed a production conversational AI chatbot on IBM Cloud using LLM inference APIs and contributed NLP-based ranking and retrieval components. Worked on evaluation cycles, rollout, observability, and iterative improvement using live traffic.
2021 - 2025
B.Tech, Computer Science & Data Science
Aug 2021 – Jul 2025
CGPA: 8.05/10. Coursework included Machine Learning, Deep Learning, Probability & Statistics, Linear Algebra, and Data Structures.
Novartis Hackathon 2025 project using an XGBoost ensemble with medCPT transformer embeddings, TF-IDF features, and SHAP explainability. Achieved MAE of 11.46 and placed in the top 12 nationally across India. Communicated model behavior, feature importance, and reliability to non-technical judges.
Personal production-grade multi-agent RAG system built with LangChain, LangGraph, ChromaDB, and FAISS. Included agent-level ranking, confidence scoring, hallucination prevention, FastAPI deployment, and stakeholder dashboards. Achieved semantic similarity above 0.87 on multilingual Indic documents.
Published research project using a CNN-RNN hybrid on the UCF-Crime dataset with image captioning and predictive risk scoring, optimized for low-latency scalable real-time inference in PyTorch.