CV
The curriculum vitae of Shubhro Dev. Use the button above to download a PDF copy.
Contact Information
| Name | Shubhro Dev |
| Professional Title | AI/ML Researcher |
| shubhro2004@gmail.com | |
| Phone | +91 8272979689 |
Professional Summary
AI/ML researcher and Computer Science undergraduate at RGIPT working on large language models, multimodal deep learning, federated learning, differential privacy and time-series analysis. Currently a Technical Research Intern at Siemens.
Experience
-
2026 - 2026 Bengaluru,
Karnataka,
IndiaTechnical Research Intern, Global DAI R (SDE-IN Team)
Siemens
- Researched validation and verification of outputs generated by LLMs.
- Experimented with deterministic ways to validate and verify code and NLP use cases.
- Evaluated Bidify (an existing RAG application) by inspecting each individual pipeline independently.
-
2025 - 2025 Jodhpur,
Rajasthan,
IndiaResearch Assistant (Supervisor: Dr. Debasis Das)
VANET Lab, IIT Jodhpur
- Studied time-series analysis, transformers, federated learning, membership inference attacks (MIA), auto-encoders, VAEs, LSTM, GRU, RNN, Bi-RNN, predictive maintenance (PdM), prognostics and health management (PHM) and differential privacy.
- Developed state-of-the-art architectures with multimodal features for RUL prediction and industrial PdM using the IDA-2024 Challenge SCANIA-X dataset.
- Proposed a novel hybrid architecture fusing a custom transformer-based time-series encoder with a sophisticated transformer, combined with differential privacy.
-
2025 - 2025 Remote
San Jose, USAData Science Intern
Core-AI Solutions
- Engineered multimodal AI models integrating non-invasive MRI neuroimaging, blood plasma biomarkers and cerebrospinal fluid parameters to predict Alzheimer’s disease onset and progression with high accuracy (AUC > 0.90).
- Conducted longitudinal predictive modeling on cohorts with 4+ years of follow-up to identify mild cognitive impairment patients at highest risk of conversion to Alzheimer’s dementia.
- Integrated explainable-AI tools (SHAP and LIME) to visualize biomarker contributions and generate clinician-friendly reports of top-ranked risk factors.
-
2024 - 2025 Dahanu,
Maharashtra,
IndiaMachine Learning Intern
Adani Power
- Built time-series forecasting models (LSTM, GRU, Prophet) to predict power generation and load demand, and analysed plant efficiency at Dahanu Thermal Power Station (DTPS).
- Applied regression analysis, correlation studies and linear programming for power-loss minimisation, with feature engineering (lag features, rolling statistics, seasonal decomposition) for transmission efficiency.
- Created Power BI dashboards tracking plant availability, heat rate, fuel efficiency and transmission losses across voltage levels (11kV, 415V).
-
2024 - 2024 Varanasi,
Uttar Pradesh,
IndiaResearch Intern (Supervisor: Dr. Sanjay Kumar Singh)
VCA Lab, IIT (BHU) Varanasi
- Studied CNNs, Vision Transformers (ViT), federated learning, transfer learning, multimodal models and knowledge distillation.
- Conceptualised a federated learning algorithm for early diagnosis of Alzheimer’s disease using multimodal biological-marker and image data.
-
2024 - 2024 Kolkata,
West Bengal,
IndiaResearch Intern (Supervisor: Dr. Ram Sarkar)
CMATER Lab, Jadavpur University
- Studied supervised learning, data preprocessing, attention mechanisms, deep neural networks and transfer learning.
- Experimented with attention, PCA, colour-channel splitting and image partitioning for lung cancer classification.
- Proposed a novel transfer-learning and attention-based architecture to detect and classify lung cancer from CT scans and histopathological images.
Education
-
2022 - 2026 Jais,
Uttar Pradesh,
IndiaBachelor of Technology
Rajiv Gandhi Institute of Petroleum Technology (RGIPT)
(An Institute of National Importance along the lines of the IITs)Computer Science and Engineering CPI 7.82
-
2020 - 2022 Kolkata,
West Bengal,
India -
2018 - 2020 Kolkata,
West Bengal,
India
Publications
-
2026 A Multiview Multiobjective Clustering Approach to Unsupervised Extractive Summarization
Target venue AACL 2026
Manuscript in preparation.
-
2025 A Differentially Private Hybrid Transformer-Mamba Architecture for Industrial Predictive Maintenance on Multivariate Time Series Data
IEEE Transactions on Intelligent Transportation Systems
In submission.
-
2025 Lung Cancer Identification from CT Scans using a Soft-attention enabled Deep Transfer Learning Model
IEEE ISACC 2025
Published at IEEE ISACC 2025 (pp. 254-259). DOI: 10.1109/ISACC65211.2025.10969319.
Projects
-
Cross-Industry Privacy-Preserving Framework for Predictive Maintenance (IDA 2024 Challenge)
A federated, differentially private framework for RUL prediction on the SCANIA-X dataset. Combines numerical and categorical features into transformer embeddings with a TabTransformer, applies Spectral-DP and DP-SGD, and aggregates a global model across heterogeneous clients with the flwr library.
- Implemented differential privacy from scratch (Spectral-DP and DP-SGD, including a Renyi DP / Moments accountant) to protect training data, with a two-stage model pairing a Transformer time-series encoder and a TabTransformer head.
- Infused federated training via the flwr library, aggregating a global model across heterogeneous clients with different computing power, parameters, and hyperparameters to mirror real-life industrial equipment.
- Best RUL prediction model on the dataset with an MSE of 2725 while preserving training-data privacy, achieving 49.12% AUC and 49.59% accuracy in an advanced membership inference attack.
-
Context-Aware Multimodal Knowledge Retrieval System
A context-aware parser that automatically extracts and separately processes multimodal content (images, tables, equations, graphs, text) and summarises each modality into vector embeddings, then answers queries with comprehensive, source-cited responses referencing text, table data and image insights.
- Employed modality-specific pipelines and selected the appropriate LLM per content type to improve summary quality, storing embeddings and summaries in ChromaDB using Hugging Face embeddings.
- Runs three parallel retrieval pipelines (LLM-summary embeddings, raw-atomic content, and CLIP text-to-image search), parsing documents with Docling.
- Designed a multi-vector retrieval strategy linking document summaries to original content, so queries retrieve the relevant multimodal content for context preservation.
-
SENTRAL - Multi Spectrum Stock Analysis using custom LLMs
A multi-spectrum stock analysis system combining fundamental analysis (14 metrics such as P/E, P/B, Debt plus Piotroski F-Score and Altman Z-Score with sector peer comparisons), technical analysis (18 indicators such as EMA20/50), multi-LLM news sentiment, and Transformer/LSTM forecasting to compute buy/hold/sell probabilities.
- Collected news via seven APIs plus web scraping, filtered items by relevance score, then fed the curated corpus to 10 LLMs (finance-tuned and SOTA reasoning models) for sentiment extraction and signal generation.
- Applied Transformer and LSTM models to forecast price trends and ensembled sentiment with fundamental and technical indicators, boosting a newbie portfolio by +45% over five months.
Awards
-
2025 Global Rank 10, Meta Hacker Cup 2025 (AI Track)
Meta
-
2025 Reviewer, International Conference on Cyber Security and Artificial Intelligence (ICCSAI 2025)
ICCSAI
-
2023 Finalist, DPBH 2023
Ministry of Consumer Affairs, Government of India
-
2024 CodeChef Max Rating 1622 (3 star)
CodeChef
-
2024 Secretary, OWASP RGIPT Student Chapter
OWASP
-
2022 Qualified JEE-Advanced 2022 (AIR 17078)
IIT / JEE
-
2022 AIR 1859, WBJEE 2022
WBJEE Board
-
2022 AIR 1899, MHT-CET 2022
Government of Maharashtra
Certificates
- Summer of Quant 2024 - IIT Kharagpur (2024)
- The Web Developer Bootcamp 2024 - Udemy (2024)
- Deep Learning Specialization - DeepLearning.AI
- Machine Learning Specialization - DeepLearning.AI
Skills
Languages: Python, C++, C, SQL, HTML, CSS
ML / DL Libraries: PyTorch, TensorFlow, Scikit-learn, Hugging Face, NLTK, OpenCV, NumPy, Pandas, Seaborn, Matplotlib, SciPy
LLMs and Agents: LangChain, LangGraph, RAG, Fine-tuning, Transformers, NVIDIA NIM
Specializations: Federated Learning, Differential Privacy, Time-Series Analysis, Computer Vision, NLP
Cloud and Tools: AWS Bedrock, S3, Lambda, SageMaker, Streamlit, Power BI, Tableau
Languages
English : Fluent
Hindi : Fluent
Bengali : Native
Marathi : Conversational
Assamese : Conversational
Interests
Research: Large Language Models, Multimodal Learning, Privacy-Preserving ML, Predictive Maintenance
Beyond Work: Chess, Building Tools, Gaming