Founding Engineer at Prudentbit
I'm an AI Engineer passionate about building intelligent systems and secure platforms. I specialize in developing and deploying scalable ML/NLP models, constructing end-to-end data pipelines, and creating AI-powered web applications. At Prudentbit, I lead efforts to integrate LLMs into real-time document analysis and secure file-sharing workflows.
- Languages: Python, Java, Rust, Bash, SQL
- AI/ML: Transformers, RAG, Transfer Learning, Scikit-learn, TensorFlow, PyTorch, Hugging Face, OpenCV
- Frameworks & Libraries: FastAPI, Django, DRF, Streamlit, LangChain, Pandas, NumPy, Matplotlib, Librosa
- DevOps & Cloud: AWS (EC2, S3), Azure, GitHub Actions, Nginx, Certbot, CI/CD, SSL/TLS
- Database: MySQL, PostgreSQL
- Concepts & Tools: Data Privacy, PII Detection, Secure Architectures, Git, Postman, VSCode, CUDA
- Led development of full-stack applications aligned with business goals.
- Designed secure, high-availability infrastructure using Nginx and automated SSL with Certbot.
- Optimized cloud deployment, cutting costs by 40%.
- Integrated Collabora WOPI for real-time collaborative document editing.
- Developed a RAG-based conversational AI with LangChain, OpenAI, and ChromaDB.
- Built multi-format document pipelines with LLM agents and tabular data understanding.
- Engineered PII-safe NLP flows supporting multiple languages (English, Arabic, Devanagari).
- Designed custom ID detection logic (e.g., Aadhaar, PAN) and context-aware prompt strategies.
- Applied LDA and BERT Topic Modeling for large-scale text data.
- Performed NER and sentiment analysis for research insights.
- Collaborated on data visualizations and findings dissemination.
- Built a CNN using TinyVGG architecture on FashionMNIST.
- Custom training/evaluation loops, accuracy tracking, and PyTorch model saving.
- Demonstrated strong understanding of deep learning workflows.
- Developed an unpaired image translation system using CycleGAN (PyTorch) on Kaggle Photo/Monet dataset (~8K photos, ~1.2K paintings).
- Implemented ResNet-based generators and PatchGAN discriminators for high-quality artistic style transfer.
- Trained for 30+ epochs with checkpointing/resume functionality and optimized data throughput using multi-worker dataloaders.
- Generated and curated a gallery of Monet-style outputs, demonstrating successful domain adaptation and artistic translation.
- LinkedIn: Manas Bisht
- Email: manasbisht2507@gmail.com
- X: manasbisht099