Available for opportunities

D.P. Sai Manohar

Data Scientist building end-to-end ML pipelines on real-world data. M.E. in AI/ML · Data Science Intern at Ericsson Global · 5th globally in the 2025 LWM Challenge.

View Projects Download Resume

5th

Global LWM 2025

ML Projects

Certifications

01 / About

Who I Am

I'm a Data Scientist with hands-on experience building end-to-end ML pipelines on real-world datasets — from EDA and feature engineering to model deployment, monitoring and CI/CD automation.

Currently pursuing M.E. in AI/ML at Manipal School of Information Sciences while interning at Ericsson Global, where I work on 6G network intelligence using Transformer encoders and LLM optimization.

I like going beyond the obvious. Most people apply SMOTE and move on. I benchmark it against four strategies across five classifiers and run statistical tests before drawing conclusions. That gap between what's assumed and what's true is where I spend most of my time.

🏆

5th Place — LWM Challenge 2025

Global competition for Large Wireless Model for 6G Spatial Intelligence. Optimized task-specific heads for simultaneous beam management and high-precision user tracking.

🎓

M.E. in AI/ML — Manipal

GPA 8.04/10. Coursework in Machine Learning, Deep Learning, Statistical Learning, NLP and Computer Vision.

📜

Deep Learning Specialization

DeepLearning.AI certification. Also certified in Fundamentals of AI Agents Using RAG and LangChain by IBM.

02 / Skills

Technical Stack

Machine Learning

XGBoost LightGBM Random Forest Scikit-learn SMOTE Logistic Regression

Deep Learning & Forecasting

PyTorch LSTM Transformers Prophet Sequence Modeling

MLOps & Deployment

MLflow FastAPI Docker GitHub Actions Streamlit Evidently AI

Data & Cloud

BigQuery GCP SQL Looker Studio Pandas NumPy

Analysis & Visualization

SHAP Matplotlib Seaborn Plotly Tableau SciPy

Languages & Tools

Python SQL R Java Git Jupyter

03 / Experience

Where I've Worked

Ericsson Global

June 2025 — Present

Data Science Intern

Architected a 12-layer Transformer encoder using SSL and Masked Channel Modeling to generate universal channel embeddings for 6G networks.
Implemented TimeLLM to reprogram frozen Llama-7B for wireless time-series forecasting, achieving 37% reduction in MSE via Prompt-as-Prefix engineering.
Secured 5th place globally in the 2025 LWM Challenge for 6G Spatial Intelligence by optimizing task-specific heads for simultaneous beam management and user tracking.

04 / Projects

Things I've Built

001

ML Imbalance Benchmark

5 classifiers × 4 sampling strategies across 3 datasets. Every result goes through Wilcoxon and Friedman tests. Finding: classifier choice matters more than SMOTE.

Python XGBoost PostgreSQL Streamlit Docker

View on GitHub

002

Real-Time Financial Fraud Detection

XGBoost on IEEE-CIS dataset (590K transactions, 27.6:1 imbalance). Threshold tuning at 0.83 achieved F1: 0.6618, ROC-AUC: 0.9485 (+55% F1). Full MLOps with drift monitoring.

XGBoost MLflow FastAPI Docker Evidently AI

View on GitHub

003

Multi-Model Demand Forecasting Engine

Forecasting on M5 Walmart dataset (58.3M records, 30,490 SKUs). Compared Prophet, LSTM and XGBoost on 28-day horizon. Prophet best RMSE: 218, XGBoost best MAPE: 5.88%.

Prophet PyTorch LSTM XGBoost Streamlit

View on GitHub

004

End-to-End Sales Forecasting Pipeline

Queried 124,903 orders from GCP BigQuery across 4 tables, engineered 22 features. XGBoost MAPE 9.25%. Automated pipeline deployed as live Looker Studio dashboard.

BigQuery XGBoost SQL Looker Studio GitHub Actions

View on GitHub

06 / Contact

Let's Talk

Open to full-time roles, research collaborations, and interesting problems. If you're working on something that involves real data and actual statistical rigor, I want to hear about it.

Send an Email

✉

manohardp01@gmail.com