Vikas Kumar

Data Scientist

Kolkata, West Bengal, India3 yrs 9 mos experience

AI ML PractitionerAI Enabled

Key Highlights

Expert in developing production-ready AI solutions.
Proficient in cloud platforms for scalable AI deployment.
Strong command of modern deep learning frameworks.

Stackforce AI infers this person is a Data Scientist specializing in AI solutions for EdTech and Document Processing.

Contact

Skills

Core Skills

Generative AiNatural Language ProcessingMachine LearningComputer VisionData Science

Other Skills

Data Wrangling & Feature EngineeringModel Deployment & MLOpsPython (Programming Language)AI Agent DevelopmentModel Evaluation & Hyperparameter TuningSQL & Structured Data AnalysisObject Detection (YOLOv3)Image Classification (ResNet)BERTCloud Platforms (Azure, GCP, AWS)Prompt Engineering & RAG SystemsDeep Learning FrameworksGenerative AI & LLM DeploymentComputer Vision & OCR IntegrationMachine Learning Algorithms

About

Hello! I am Vikas Kumar, a passionate Data Scientist with over 3 years of hands-on experience in Machine Learning, Natural Language Processing, GenerativeAI, Deep Learning, and Computer Vision. My expertise lies in developing production-ready AI solutions that drive efficiency and innovation. I have a strong command of modern deep learning frameworks such as PyTorch and TensorFlow, which I leverage to create and deploy Generative AI applications. My journey in AI has been fueled by a desire to solve real-world problems and contribute to impactful projects. Throughout my career, I have successfully designed and implemented advanced AI systems, including personalized tutoring platforms and document processing workflows. I am proficient in cloud platforms like Azure, GCP, and AWS, which allows me to deploy scalable AI solutions effectively. My technical skills are complemented by my ability to collaborate with cross-functional teams and mentor junior professionals. I am always eager to learn and stay updated with the latest trends in AI and machine learning.

Experience

3 yrs 9 mos

Total Experience

3 yrs

Average Tenure

8 mos

Current Experience

Deloitte

Data Scientist

Sep 2025 – Present · 8 mos

Aimunshi- high performance enterprise automation

2 roles

Data Scientist

Promoted

Apr 2023 – Nov 2024 · 1 yr 7 mos · Hyderabad, Telangana, India

1) Personalized Tutor RAG (GenAI App)
Designed and implemented a Retrieval-Augmented Generation (RAG) system using LangChain, FAISS, and OpenAI GPT-4, enabling personalized tutoring via natural language queries over multi-subject academic notes.
Achieved high retrieval precision by incorporating token-based chunking with overlapping windows and metadata enrichment (subject, page number) to effectively index 500+ page PDFs per subject.
Integrated subject-specific filtering and GPT-4 prompting logic, ensuring domain-relevant, hallucination-free responses in disciplines like DBMS, OS, DSA, C Programming, and Algorithms.
Engineered a chat interface with session history, enabling contextual multi-turn interactions and mimicking an intelligent tutor experience via Streamlit UI.
Enabled vector persistence and cold-start efficiency by implementing one-time FAISS vector generation logic and fast reload during subsequent app runs.

Data Wrangling & Feature EngineeringModel Deployment & MLOpsGenerative AINatural Language Processing

Associate Data Scientist

Sep 2021 – Mar 2023 · 1 yr 6 mos · Hyderabad, Telangana, India

1)AIMunshi - Document Extraction Tool (LayoutLM)
Developed a BERT-based LayoutLM model, improving document classification accuracy to 93%.
Implemented Google OCR and Tesseract OCR, increasing text extraction efficiency by 40%.
Created the final result from the predictions to form a structured result using Pytorch.
Deployed the model on Azure and GCP, reducing inference time by 30%.
Gained knowledge of Encoders, Decoders, and the Attention mechanism.
2) Invoice Orientation Correction (Image Classification)
Built a ResNet18-based orientation detection model, achieving 98% accuracy.
Trained the model to identify the orientation of the invoices from Straight, Inverted, Clockwise, and Anti-clockwise.
Integrated model into OCR pipeline, reducing error rates by 25%.
3) Signature and Stamp Detection (Object Detection)
Trained a YOLOv3 model, achieving 95% accuracy in detecting invoice verification marks.
Reduced manual verification workload by 40%, accelerating invoice processing.