[Software Services] : Sr. AI/ML Developer

Location: Noida, India

Department: Software Delivery

Experience: 6 to 8 Years

• Design, develop, and optimize end-to-end machine learning models for classification, regression, NLP, computer vision, and recommendation tasks.
• Conduct feature engineering, model selection, hyperparameter tuning, and performance evaluation using industry-standard frameworks.
• Ensure models meet production-grade accuracy, latency, and scalability requirements.

• Architect and implement scalable ML training pipelines with robust data ingestion, preprocessing, augmentation, and validation stages.
• Design and manage distributed training workflows (single-node and multi-GPU/TPU clusters) using frameworks such as PyTorch, TensorFlow, and JAX.
• Build automated experiment tracking and reproducibility systems using tools like MLflow, Weights & Biases, or DVC.

• Deploy, monitor, and maintain ML models in production using containerized (Docker, Kubernetes) and cloud-native infrastructure.
• Establish CI/CD pipelines for model training, evaluation, versioning, and automated re-training triggers.
• Implement model performance monitoring, drift detection, and alerting systems to ensure sustained model health.
• Manage model registries and artifacts using platforms such as SageMaker, Vertex AI, Azure ML, or open-source MLOps stacks.

• Build and deploy production-grade Generative AI applications using leading platforms and open-source models.
• Proprietary: OpenAI GPT-4/o, Anthropic Claude, Google Gemini, Amazon Titan, Cohere
• Open-source: Meta LLaMA 2/3, Mistral, Falcon, Mixtral, Phi-3, Gemma
• Design prompting strategies including zero-shot, few-shot, chain-of-thought (CoT), and structured output prompting for diverse task types.
• Evaluate LLM outputs for hallucination, toxicity, relevance, and faithfulness using automated and human-in-the-loop evaluation frameworks.

• Design and implement Retrieval-Augmented Generation (RAG) pipelines with semantic chunking, hybrid search (dense + sparse), reranking, and document parsing strategies.
• Build and orchestrate Agentic AI systems with memory, planning, tool use, and multi-agent collaboration using LangChain, LangGraph, AutoGen, CrewAI, or similar frameworks.
• Apply parameter-efficient fine-tuning (PEFT) techniques, particularly Low-Rank Adaptation (LoRA) and QLoRA, for domain adaptation and instruction tuning of large language models.
• Implement Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) pipelines where applicable.

• 6–8 years of professional experience in machine learning and AI development.
• Strong proficiency in Python and ML libraries: NumPy, Pandas, Scikit-learn, PyTorch, and/or TensorFlow.
• Proven experience designing and training ML models across supervised, unsupervised, and self-supervised learning paradigms.
• Hands-on experience building and maintaining ML training and inference pipelines at scale.
• Solid MLOps experience: containerization, orchestration, CI/CD, model monitoring, and cloud deployment (AWS, GCP, or Azure).
• Demonstrated experience with at least two major Generative AI platforms (proprietary or open-source).
• Deep understanding of Transformer architecture and large language model internals.
• Practical experience implementing RAG systems with vector databases (Pinecone, Weaviate, ChromaDB, Qdrant, pgvector, Milvus etc.).
• Hands-on experience with Agentic AI frameworks and multi-step reasoning pipelines.
• Working knowledge of Low-Rank Adaptation (LoRA/QLoRA) and other PEFT fine-tuning methods.
• Strong understanding of evaluation frameworks for both discriminative and generative models.

• Experience with multimodal models (vision-language, text-to-image, speech-to-text).
• Familiarity with model quantization techniques (GPTQ, AWQ, bitsandbytes) for efficient inference.
• Exposure to graph neural networks, time-series forecasting, or reinforcement learning.
• Contributions to open-source ML or GenAI projects.
• Experience with advanced vector search strategies: ColBERT, hybrid BM25 + ANN, reranking models.
• Knowledge of responsible AI principles, bias mitigation, and model safety evaluation.
• Publications or patents in machine learning or AI domains.