Pranshu Chaurasia | Data Scientist & AI Researcher

About

AI Researcher

I'm a Data Scientist at SocGen AI, working on Responsible and Applied AI Research. With 4+ years of experience, I design and deploy production-grade AI systems for enterprise applications in banking and finance.

My work focuses on LLM fine-tuning, RAG pipeline development, and AI evaluation frameworks. I specialize in building multi-agent workflows, custom evaluation metrics, and document processing systems that meet regulatory compliance standards.

Passionate about bridging AI research with real-world impact, I'm open to collaborating on projects involving Generative AI, Responsible AI governance, and scientific research.

Generative AI AI Evaluation & Safety Scientific Research AI Research Document Intelligence

Experience

Data Scientist – Responsible and Applied AI Research — SocGen AI

Feb 2025 – Present

Part of SocGen AI, a dedicated entity within the Group focused on accelerating AI integration to automate and optimize banking processes. Working on responsible AI deployment, regulatory compliance (EU AI Act), and building innovative solutions with strong emphasis on operational efficiency.

Data Scientist – AIS — Société Générale

2023 – 2025

Built end-to-end Document Intelligence solutions including automated document validation pipelines, on-premise RAG systems for financial documents, and LLM-based compliance assessment tools.

Big Data Engineer — Société Générale

2021 – 2023

MATLAB Intern — PANTECH SOLUTIONS Pvt. Ltd.

Sep 2020

Projects

Image Indexing & Retrieval with Qdrant

Efficient image indexing and retrieval using Qdrant with multimodal models including ColPali, ColQwen, VDR-2B-Multi-V1, and Jina embeddings v4. Enables scalable semantic search across large image datasets.

QdrantColPaliMultimodal SearchVector DB

View on GitHub

Automated Industrial Robots Simulation

Simulation of automated industrial robots in cylinder liner production lines demonstrating 73.42% improved efficiency and 34.78% reduction in manufacturing time compared to manual processes.

RoboticsSimulationIndustrial Automation

View IEEE Paper

Low-Cost Humanoid Head Design

Developed two novel humanoid heads (Flat Affect & Symbolic) with a unique 3-servo eye mechanism achieving 71.76% volume reduction and 46.80% power savings. Integrated EmotionNet on Jetson Nano for real-time facial analysis.

RoboticsComputer VisionJetson NanoEmotionNet

View IEEE Paper

Publications

IEEE CCWC 2023

Symbolic and Flat Affect Humanoid Head Design for Modular Humanoid Robots

P. Chaurasia, R. A. Viji, A. P. Chavan, et al.

Novel 3-servo eye mechanism achieving 71.76% volume reduction and 46.80% power savings compared to traditional 6-servo designs. Developed two distinct humanoid heads with modular integration capabilities.

📄 View Paper

IEEE ASIANCON 2021

Design and Implementation of Automated Industrial Robots in Cylinder Liner Production Lines

P. Chaurasia, R. Antony Viji, et al.

Simulation of automated industrial robots demonstrating 73.42% improved efficiency and 34.78% reduction in manufacturing time compared to manual processes.

📄 View Paper

Skills

Languages

PythonSQL

NLP / Generative AI

RAG PipelinesLLMsPrompt EngineeringFine-tuning (PEFT, LoRA, GRPO)Unsloth

Frameworks & Libraries

PyTorchKedroLangChainLlamaIndexLangGraph

Computer Vision

Document Layout AnalysisOCRObject DetectionVision TransformersCNN

Cloud & MLOps

Azure AI StudioDockerGit

Databases

QdrantWeaviateFAISSPostgreSQLMongoDB

AI Evaluation & Safety

G-EvalAdversarial TestingHallucination DetectionMRMEU AI Act Compliance

Core Skills

StatisticsScientific ResearchRLHFSupervised LearningUnsupervised Learning

Hi, I'm Pranshu Chaurasia