Baqir Sheikh

About Baqir Sheikh

I\'m an AI Engineer passionate about building intelligent systems that solve real-world problems through Machine Learning, Natural Language Processing, and Large Language Models.Currently working at the Center for the Development of Advanced Computing (C-DAC), I specialize in developing multilingual AI solutions that bridge language barriers. I\'ve built spam detection and sentiment analysis systems across English, Hindi, Urdu, and Punjabi with 95%+ accuracy, and created zero-shot classification tools using transformer architectures like BART and RoBERTa.My expertise spans:Generative AI & LLMs – Building RAG systems with LangChain, implementing GraphRAG with Neo4j, and reducing hallucinations by 78% through multi-agent architecturesNLP & Transformers – Fine-tuning models (DeBERTa, RoBERTa, BART) for classification, sentiment analysis, and suggestion miningComputer Vision – Developing real-time detection pipelines with YOLOv11 and custom U-Net architectures for autonomous systemsMultilingual AI – Creating inclusive solutions for underrepresented languages with custom language identification modulesI thrive on tackling challenging problems in noisy, domain-specific datasets and optimizing models for production deployment. Whether it\'s few-shot learning, hybrid retrieval systems, or graph-based knowledge extraction, I\'m committed to pushing the boundaries of what AI can achieve.Currently pursuing my B.Tech in Artificial Intelligence from Aligarh Muslim University (CGPA: 8.19/10), I\'m always excited to collaborate on innovative AI projects that create meaningful impact.

Experience

1. Ai Ml Engineer

   Cdacindia

   Jun 2023 — Present · Pune, IN

   Built a zero-shot content categorization tool using Facebook’s BART model, achieving 82% accuracy and 0.79 F1-score without labeled training data by leveraging bidirectional and auto-regressive learning.• Achieved 91.6% accuracy and 0.91 macro F1-score on multilingual tweet sentiment analysis by training RoBERTa modelswith custom language detection modules (93.4% accuracy).• Enhanced model performance on noisy, domain-specific datasets by 18% by applying few-shot learning techniques tofine-tune transformers with minimal training examples (10–50 samples per class).• Developed a multilingual spam detector in English, Hindi, Urdu, and Punjabi, achieving over 95% accuracy in English,Hindi and Urdu & 92% in Punjabi by addressing script-specific challenges Devanagari, Nastaliq), implementing Unicode normalization, and fine-tuning BERT-based models on domain-specific spam patterns.

Education

• Jaypee University Anoopshahr

  High School

• Aligarh Muslim University

  Bachelor of Technology - BTech, Artificial Intelligence

• Aligarh Muslim University

  Senior Secondary School Certificate , Physics , Chemistry , Maths