Santosh Rana - ML & Data Science Portfolio

01 / About

Building ML systems that work in the real world

Hi, I'm Santosh, a Machine Learning and Data Science engineer focused on turning strong model ideas into systems that can actually run in production.

My recent work spans landmark-based computer vision, multilingual aspect-based sentiment analysis, and real-time inference pipelines built for practical use instead of benchmark-only demos.

I care about efficient architectures, clean preprocessing, honest evaluation, and the details that make ML systems stable outside the notebook.

Explore my GitHub

02 / Projects

Selected work across vision and language

PROJECT 001

Real-Time Multimodal CV System

A dual-pipeline webcam system combining hand gesture recognition and facial emotion recognition using MediaPipe landmarks, a lightweight MLP, and a GCN-based facial classifier.

              95%
              Gesture Accuracy
            

              20-30
              FPS Live
            

              19K
              MLP Params
            

Focus: Landmark-driven real-time interaction without heavy CNN pipelines, including dynamic wave detection and geometry-aware facial emotion modeling.

PyTorch MediaPipe GCN OpenCV Computer Vision

Open Project Open Code

PROJECT 002

Korean ABSA Benchmark

An end-to-end benchmark comparing Classical ML (TF-IDF + LR), a Korean Transformer (KcELECTRA), and a small LLM (Qwen 2.5) for aspect-based sentiment analysis on Korean restaurant reviews.

              0.94
              Best Mention F1
            

              3
              Models Tested
            

              4
              Aspects
            

Focus: Benchmarking trade-offs between performance, model size, and deployment practicality for FOOD, PRICE, SERVICE, and AMBIENCE sentiment understanding in Korean. KcELECTRA leads overall; TF-IDF + LR is the strongest lightweight option.

NLP KcELECTRA Qwen 2.5 TF-IDF Korean Benchmarking

Open Project Open Code

03 / Skills

Core stack for research, training, and deployment

ML Engineering

PyTorch
PyTorch Geometric
Scikit-learn
AdamW / optimizers
Model evaluation

Computer Vision

OpenCV
MediaPipe
Landmark pipelines
Real-time inference

Natural Language

Transformers
HuggingFace
Aspect sentiment analysis
Benchmark design
Prompt evaluation

LLMs & Deployment

Ollama
Streamlit
Few-shot prompting
Deployment-minded prototyping

Workflow

Python
Data preprocessing
Experiment tracking
Git

04 / Contact

Let's build something useful

If you want to discuss ML systems, research implementation, or practical deployment work, I'm happy to connect.

Email Me Download CV ↓

GitHub LinkedIn