Saurabh Shrivastava
Data Scientist with 5+ years of R&D experience across safety-critical systems at ISRO and real-time ML deployment at Mercedes-Benz. Trained in applied machine learning through an M.Tech in AI from IISc Bangalore. Experienced in designing, training, and deploying production-grade ML systems for computer vision and data-driven decision-making.
Experience
Data Scientist
Mercedes-Benz R&D
Deployed real-time computer vision pipelines for in-cabin monitoring on automotive edge platforms. – Exposure Quality Assessment: Engineered a multi-level detection pipeline for in-cabin monitoring; leveraged VLMs for automated large-scale labeling and deployed a lightweight tree classifier, filtering out 7% of poorly exposed data. – Image Denoising for Eye-Gaze: Mitigated noise in degraded eye images via a gaze-aware denoising pipeline; synthesized Ground Truth using Diffusion Models and trained a CNN based inference model. Reduced gaze error by 0.3°. – Head Orientation Modules: Developed and deployed CNN-based Head Orientation (HO) modules for Driver Monitoring Systems (DMS) across 3 carlines; achieved a Mean Absolute Error (MAE) of <2.4° and a 2-sigma error bound of <7.1° for Yaw, Pitch, and Roll, ensuring high-reliability inference in production.
Data Science Intern
Udaan
– Causal Attribution Modeling: Quantified the incremental contribution of ad channels using Causal Inference with probabilistic models on observational data to optimize marketing spend. – Experimentation: Designed and executed hypothesis-driven A/B tests to validate budget allocation strategies, leading to improved decision-making around Return on Ad Spend (ROAS).
Scientist/Engineer "SC"
ISRO
– Owned and maintained safety-critical software modules for launch vehicles (PS4, GSLV MK2/MK3), supporting Liquid Stage Servicing systems. – Collaborated with Software, Data Science, and Operations teams across 13 successful launch missions, operating under strict reliability and real-time constraints. – Impact: Discovered and fixed a latent defect in a flow regulation module (failure probability 0.016) active in production for 12 years, improving system safety and mission robustness.
Education
Indian Institute Of Science, Bangalore
M.Tech
CGPA: 8.20
National Institute of Technology, Raipur
B.Tech
CGPA: 9.04
Skills
PROJECTS
Addressing Domain Shift in Echocardiography Segmentation | Master’s Project
– Developed EchoSAM-Flex to address domain shift in medical image segmentation; conducted systematic benchmarking against state-of-the-art segmentation models across CAMUS and EchoNet-Dynamic datasets. – Defined cross-domain generalization metrics and evaluated robustness under distribution shift, achieving relative Dice improvements of 44.07% and 16.65% without additional annotations.
Session-based Apparel Recommendation System (Winner) | IISc Datathon
– Designed an exploitation–exploration strategy using a decayingε-greedy policy for session-based recommendations. – Used pre-trained MobileNetV2 embeddings with exponential smoothing (exploitation) and random sampling (exploration).
Weather Prediction for Launch Operations | ISRO, Sriharikota
– Developed LSTM and GRU models to forecast multiple weather parameters for launch-day operational decision support. – Improved pressure prediction accuracy by reducing MAE from 2.5 to 0.32 hPa (87% reduction) over existing methods.
Abstractive Text Summarization using Transformers | Course Project
– Implemented Encoder-Decoder architecture with Bahdanau Attention on the Amazon Fine Food Reviews dataset. – Conducted comparative evaluation of transformer-based summarization models; analyzed performance using BLEU and qualitative error analysis to identify reasoning limitations and generalization gaps. Improved BLEU score by 2.55% through structured architectural refinements.