AI Engineer working across Generative AI, Large Language Models, Computer Vision, Deep Learning, Machine Learning, and NLP.
Currently contributing to the Keras ecosystem
I enjoy building production-ready AI systems and contributing to open-source ML infrastructure.
- 🧠 Generative AI & LLMs — RAG, multimodal AI, agents, evaluation and inference
- 🤗 Keras / KerasHub — model integration, checkpoint conversion, presets and numerical verification
- ⚙️ ML Infrastructure — tokenizers, preprocessing, data pipelines and multi-backend support
- 🤝 Agentic AI — Multi-Agent Systems, Google ADK, A2A and MCP
- 👁️ Computer Vision & Edge AI — object detection, OCR, segmentation and real-time video analytics
- 🟢 NVIDIA AI Stack — NVIDIA DeepStream, NVIDIA TAO Toolkit, NVIDIA Warp and TensorRT
- ☁️ Cloud & MLOps — GCP, AWS, MLflow, Docker and CI/CD
I actively contribute to the Keras ecosystem, working on AI/ML infrastructure, model integrations, checkpoint conversion, preprocessing, tokenization, numerical parity and multi-backend support.
Building reliable ML infrastructure and bringing modern AI models into production-ready Keras workflows.
Python SQL C C++
Generative AI LLMs NLP RAG Machine Learning
Deep Learning Computer Vision Multimodal AI
Object Detection OCR Image Segmentation
Keras KerasHub TensorFlow PyTorch
Hugging Face ONNX TensorRT OpenCV
Scikit-learn LangChain LlamaIndex
LiteLLM LangGraph YOLO Detectron2
ModernBERT RoBERTa Qwen FLUX
Gemma ViT CNN RNN GAN U-Net YOLO
Google ADK A2A MCP
Multi-Agent Systems Agentic RAG Text-to-SQL
GCP Vertex AI AWS Bedrock
SageMaker S3 EC2 DynamoDB
MLflow Weights & Biases Docker
CI/CD Harness
PostgreSQL MySQL MongoDB
FAISS ChromaDB Milvus OpenSearch
PEFT LoRA QLoRA
Transfer Learning Distributed Training
vLLM LLM Evaluation A/B Testing
Multi-Agent system for supply-chain operations using:
Google ADK A2A MCP PostgreSQL Agentic RAG
The system uses an orchestrator agent to delegate tasks across Inventory, Logistics and Finance agents, with MCP-based database access and approval-gated workflows for sensitive financial operations.
Ai based Face-recognition Attendance System.
Custom fashion recommendation system using an open-source Meta Llama 3 model with a RAG pipeline.
Tech: Llama 3 RAG FAISS LangChain
Uses FAISS as the vector retriever to retrieve relevant fashion information and generate personalized recommendations.
Computer-vision system for determining the distance between nearby vehicles and identifying potentially dangerous proximity.
Tech: YOLOv5 3D Euclidean Distance OpenCV
Calculates 3D (x, y, z) Euclidean distance between vehicles and tracks vehicle pairs within a defined 15-foot threshold, enabling potential CCTV-based accident detection.
Final-round project for the NVIDIA AI Hackathon 2021, featuring multiple computer-vision modules:
- 😷 Mask Detection
- 📏 Safe Distance Detection
- 🤧 Sneezing & Coughing Detection
- 📱 Phone-use Detection
🏆 Final-round project with a high ranking
Custom-trained face detection model using YOLOv7-Tiny, designed for accurate and faster inference.
Tech: YOLOv7-Tiny Custom Training Computer Vision
Explored semantic and instance segmentation using U-Net across medical, satellite and general image datasets.
Applications include:
- Medical image segmentation
- Background removal & Image replacement
- Satellite image segmentation with Smooth image blending
🎥 **View Demo
Top 10 in the final round with a computer-vision solution involving mask detection, safe-distance detection and action detection.
📧 Email: sinhamaitry6@gmail.com
💼 LinkedIn: Maitry Sinha
💻 Old personal GitHub: Maitry Sinha
Open Source · Keras · TensorFlow · Generative AI
LLMs · Agentic AI · Computer Vision · Deep Learning
ML Systems · MLOps · AI Infrastructure · Cloud . MLOps
