Dr. Niraj Bagh

Company Name: Philips HealthCare (Philips Innovation Campus)

Year: Oct 03, 2025-Present

City: Bangalore, India.

Role: Senior Research Scientist

- Responsibilities:.

As a Senior AI Scientist at Philips, I leverage deep learning, computer vision, and generative AI to revolutionize radiology by supporting doctors and healthcare professionals in making smarter, faster, and more accurate diagnoses, optimizing medical imaging workflows, reducing clinical burden, and improving patient outcomes worldwide.

Company Name: TeraRecon, Concert AI, Symphony AI

Year: Jan 01, 2025-Sep 06, 2025

City: Bangalore, India.

Role: Senior Data Scientist

- Responsibilities:
  - - Fine-tuned LLaMA 3.1 (8B) using QLoRA (8-bit quantization) for radiology report summarization on unstructured clinical text, achieving CheXbert F1 >0.80 and RadGraph F1 >0.85.
    - Built an Agentic AI system with LLaMA 3.1 (8B) to summarize radiology reports, achieving strong clinical summarization metrics (BLEU/ROUGE-L >0.33, BERTScore P/R/F1 >0.7).
    - Leveraged LLaMA 3.1 (405B) on AWS Bedrock using zero-shot and few-shot prompting, achieving high clinical correctness, completeness, and conciseness, validated by expert radiologists for accurate radiology report summarization.
    - Utilized locally hosted LLaMA 3.1 (8B), DeepSeek-R1 (7B), and Mistral (7B) for psychological disorder classification via zero-shot and few-shot prompting; LLaMA 3.1 (8B) with few-shot prompting achieved the best results (BERTScore F1 >0.43, BLEU/ROUGE-L >0.21).
    - Developed a RAG-based Q&A App on neuroscience using GPT-4, LangChain, and Chroma vector database, enabling faster retrieval of Brain-Computer Interface (BCI) queries from scientific sources.

Software: Radiology AI, Medical AI, Heathcare AI, LLM Models, Transformers, Python (Scikit-Learn, TensorFlow, Numpy, Pandas, PyTorch, PyCharm, Keras, spaCy and NLTK, OpenCV), NannyML, LLM (Hugging Face, Lang chain and Llama Index AWS (Amazon Sagemaker).

Company Name: Nanoprecise Data Service Pvt. Ltd., India.

Year: Dec 01, 2022-Oct 06, 2024

City: Bangalore, India.

Role: R & D Engineer (R & D)

- Responsibilities:
  - - Collected and analyzed IoT sensor data, enabling the transformation of industrial machines into intelligent, Industry 4.0-compliant systems.
    - Developed algorithms for assessing the health of industrial machines, contributing to predictive maintenance strategies.
    - Investigated sensor signal quality and continuously monitored real-time data from industrial equipment to ensure operational efficiency.
    - Collaborated closely with hardware, firmware, and condition monitoring teams to conduct fault analysis and improve machine reliability.

Key Projects:
- - Health Monitoring: Developed and deployed algorithms to detect drifts in sensor data from industrial machines, resulting in a 75% reduction in false positive alarms.
  - ·ML Experiment Pipeline: Designed and implemented a data processing pipeline in Kedro, encompassing data ingestion, cleaning, and machine learning model integration for fault detection and pattern recognition.
  - Few-shot Learning: Designed and deployed a Few-shot (One-shot and Five-shot) learning model to classify various faults (Healthy, Unbalance, Misalignment, Inner Race, Outer Race and Ball Bearing) in industrial machines using limited acceleration data.
  - GAN-based Image Generation: Successfully generated STFT-based spectrogram images using GANs and performed fault analysis with sophisticated 2D CNN models.
  - Fault Classification: Designed a customized 2-D CNN model and a 1-D CNN +LSTM model to classify multiple faults (Healthy, Unbalance, Misalignment, Inner Race, Outer Race and Ball Bearing) in industrial machines, enhancing overall system diagnostics accuracy by 90%.
  - Transfer Learning: Leveraged transfer learning techniques (AlexNet, VGG-16, Inception, and ResNet) to identify multiple faults in industrial machinery using time-frequency distribution-based images.
  - Signal-to Image Transformation: Implemented advanced time-frequency distribution techniques (Spectrogram, Continuous Wavelet Transform Wavelet-based Synchrosqueezing Transform, Fourier-based Synchrosqueezing Transform, Wigner-Ville Distribution, Constant-Q Nonstationary Gabor Transform, Hilbert Huang Transform, Scattergram) to generate images for training deep learning models on machine fault classification.
  - Harmonic Series Detection: Analyzed the harmonic series of fundamental and fault frequencies (BPFO, BPFI, BSF, and GMF) using Harmonic Product Spectrum, Kurtogram, and Cepstrum analysis.
  - ML Models Monitoring: Employed NannyML to monitor ML models, focusing on performance estimation and change detection, which enhanced machine lifespan prediction accuracy by 20%.

Software: Python (Scikit-Learn, TensorFlow, Numpy, Pandas, PyTorch, PyCharm, Keras, spaCy and NLTK, OpenCV), NannyML, LLM (Hugging Face, Lang chain and Llama Index AWS (Amazon Sagemaker).

Company Name: Qneuro Pvt. Ltd., India.

Year: May 17, 2021-Sep 30, 2022

City: Chennai, Tamil Nadu, India.

Role: Neuroscientist/Data Scientist

Responsibilities:
- - Led cutting-edge neuroscience projects, including real-time emotion detection, cognitive load estimation, and face emoji control using brain (EEG) signals.
  - Designed and conducted neuroscience experiments to record and decode bio-signals (EEG, EOG, ECG, EMG). Applied advanced signal processing techniques to extract significant features for analysis and interpretation.
  - Designed, implemented, and evaluated ML/DL models to decode various bio-signals.

Key Project:

Emotion Detection: Deployed ML models (SVM and NB) to detect emotions from EEG signals with 90% classification accuracy.
- - - Data acquisition and preprocessing: Recorded and processed EEG data from subjects in three emotional states (Calm, Normal, and Happy) using filtering, ICA, and ASR techniques.
    - Feature Extraction and Statistical Tests: Extracted features from EEG frequency bands and conducted statistical test, including correlation analysis, t-tests, and Cohen's d test, for feature selection.
    - ML Model: Developed, trained, and deployed machine learning models (SVM, Naïve Bayes, KNN) on MATLAB and Python, with real-time implementation in C# to accurately detect emotions.

Cognitive Load Estimation: Conducted N-back test experiments, recorded and analyzed EEG data using SVM, RF, and XGBoost models to estimate cognitive load with 92% classification accuracy.
- - - Data acquisition and preprocessing: Collected EEG data from subjects during low and high cognitive load states induced by the N-back test and applied preprocessing techniques such as filtering, ICA, and CA.
    - Feature Extraction and Statistical Tests: Extracted features from EEG frequency bands and conducted statistical analyses, including ANOVA and Principal Component Analysis (PCA).
    - ML Model: Developed, trained, and evaluated machine learning models (SVM, Naïve Bayes, Random Forest, XGBoost) on MATLAB and Python, achieving high performance.

Engagement Index Estimation: Evaluated engagement levels of users from EEG signals and deployed ML models (SVM and LR) in real-time for dynamic assessment
- - - Data acquisition and preprocessing: Recorded EEG signals to assess subjects' engagement levels (low, medium, high) during activities such as playing video games, watching videos, and solving arithmetic problems.
    - Feature Extraction and Statistical Tests: Applied preprocessing techniques (filtering, ICA, ASR) and extracted features from EEG frequency bands, followed by statistical analyses (ANOVA, PCA).
    - ML Model: Developed and deployed machine learning models (Logistic Regression, SVM) in MATLAB, Python, and C# for real-time assessment of engagement levels.

EEG-based Eye Blink Detector (Single Eye Blinks and Left/Right Eye Winks): Developed a real-time eye blink detector with a 95% accuracy rate.
- - - Data acquisition and preprocessing: Collected EEG signals during single eye blinks and left/right eye winks, applying filtering techniques (50/60 Hz notch filter and band-pass filter).
    - Feature Extraction and Statistical Tests: Extracted features related to blinks and winks, followed by statistical analysis (Correlation Analysis, t-test).
    - ML Model: Developed machine learning models (Multinomial Logistic Regression, SVM) and deep learning models (1-D and 2-D CNN), and deployed them in C# for real-time detection of eye blinks and winks.

Real-Time Motor Imagery BCI System: Developed a real-time Motor Imagery BCI system with a 90% accuracy rate that enables users to move a ball left or right on a monitor screen through imagined hand movements.
- - - Data acquisition and preprocessing: Collected EEG signals while subjects imagined left and right-hand movements.
    - Feature Extraction and Statistical Test: Applied preprocessing techniques, including filtering and Independent Component Analysis (ICA). Extracted features from EEG frequency bands and conducted statistical analyses (ANOVA, PCA).
    - ML Model: Deployed machine learning models (Logistic Regression, LDA, Naïve Bayes, SVM) for real-time Motor-Imagery BCI.

Glucose Level Estimation: Implemented a 1D CNN model to estimate glucose levels using PPG signals, resulting in improved prediction accuracy.

Software: MATLAB, C++, C#, Python (Scikit-Learn, TensorFlow, Numpy, Pandas, PyTorch, PyCharm, Keras, spaCy and NLTK, OpenCV), NannyML, AWS (Amazon Sagemaker).

Organization Name: Institute of Minerals and Materials Technology (IMMT)-CSIR, Govt. of India.

Year: 2014-2015

City: Bhubaneswar, India.

Role: Project Assistant-III (2014-2015)

Responsibilities:
- - Designed a prototype model of a ball mill machine and developed a 3-D model using image processing to investigate the wear of liners and lifters in the machine.
    - - Data Collected: The images of ball mill machine were captured through NI computer system integrated with NI smart camera.
      - Image Processing Techniques: Perspective transform, image conversion (binary to skeletonized image).
      - 3-D Surface Reconstruction: Transform pixel world to real world, interpolation and median filter.

Software: MATLAB and C++

Technical Skills

• Hardware:

Arduino Uno, OpenBCI, Raspberry Pi.

• Software:

MATLAB, Python, LabView, R-studio, C/C++, Anaconda Navigation, Google Colab, Jupyter Notebook, Visual Studio and Studio 3T.

• ML/DL Packages and Model Deployment:

Scikit-Learn, TensorFlow, Numpy, Pandas, PyTorch, PyCharm, Keras, spaCy and NLTK, OpenCV and NannyML. Flask, Amazon Sagemaker, MLFlow and Google AutoML (Vertex AI), FastAPI, Streamlit, Git, Docker, MLflow, Model Deployment, CI/CD Pipelines, Model Monitoring

• Data Visualization Tool:

Microsoft Power BI, Tableau, Pygwalker, Plotly, Matplotlib, Seaborn.

• AI Tool:

ChatGPT, Gemini, Midjourney, DALL-E2, Microsoft Copilot, RAG Systems, Prompt Engineering.

• LLMs Tool:

LangChain, Hugging Face and LlamaIndex, GPT, BERT, LLaMA, DeepSeek, Hugging Face, spaCy, NLTK, CNNs, RNN, LSTM, Transformers, Diffusion Models, GANs, VAEs, Scikit-Learn, TensorFlow, PyTorch, Keras, XGBoost, Ensemble Methods.

• Big Data:

Apache Airflow

• Databases:

MongoDB, MySQL, Chroma, FAISS, Weaviate.

• Data Science:

Statistical Analysis, Predictive Modeling, Feature Engineering, Model Validation

• Computer Vision:

OpenCV, Image Classification, YOLO, Object Detection, Transfer Learning

Page updated

Google Sites

Report abuse