AI Engineer (Agentic Systems, Deployment & Infrastructure Sizing)
AI Engineer (Agentic Systems, Deployment & Infrastructure Sizing)Job Description (JD)Role SummaryWe are looking for a specialized AI Engineer with deep expertise in architecting, building, and deployingautonomous AI Agents alongside a strong command of the underlying hardware and software infrastructurerequired to run them efficiently. Unlike purely API-bound developers, you will be responsible for the full stack:designing multi-agent workflows, implementing RAG (Retrieval-Augmented Generation) and tool-use pipelines,and sizing/optimizing the compute infrastructure (GPUs, VRAM, local/on-premise servers, and cloud instances) tobalance latency, accuracy, and operational cost.Key Responsibilities● AI Agent Development: Design, build, and orchestrate autonomous single- and multi-agent systems capableof complex reasoning, planning, memory management, and external tool/API execution (using frameworkslike LangGraph, CrewAI, AutoGen, LlamaIndex, or custom orchestrators).● Hardware Sizing & Infrastructure Architecture: Evaluate, specify, and procure/provision hardwarestacks for local, edge, or data-center AI workloads. Calculate exact GPU compute, VRAM bandwidth, CPU,RAM,storage IOPS, and thermal/power requirements for running open-weight LLMs, vision models, andembedding pipelines.● Model Deployment & Serving Optimization: Deploy and serve open-source and proprietary models usinghigh-throughput inference engines (vLLM, TensorRT-LLM, TGI, Triton Inference Server, Ollama). Applyquantization (AWQ, GPTQ, GGUF, FP8) to maximize throughput on target hardware.● RAG & Enterprise Integration: Build production-grade vector search pipelines (Milvus, Qdrant, pgvector)and securely connect AI agents to internal databases, ERPs, document management systems, andIoT/operational data streams.● Software Stack & MLOps: Containerize and orchestrate AI services using Docker, Kubernetes, and NVIDIAContainer Toolkit (CUDA/cuDNN stack management). Set up CI/CD pipelines for agent evaluation,observability (LangSmith, Arize Phoenix, Prometheus), and guardrails.● Cost & Performance Benchmarking: Continuously benchmark token generation speeds, Time-To-FirstToken (TTFT), concurrency limits, and cost-per-1k-tokens between self-hosted hardware and commercialcloud APIs.Required Qualifications & Skills● Education: B.Tech/B.E. or M.Tech/M.S. in Computer Science, Artificial Intelligence, Data Science, orequivalent practical experience.● Experience: 3–5+ years in Software/ML Engineering, with at least 1.5–2+ years specifically focused onLLMs, GenAI, and Agentic workflows in production.● Agentic & LLM Stack: Advanced Python proficiency; deep familiarity with agent orchestration frameworks,function calling, Model Context Protocol (MCP), structured outputs, and vector databases.● Infrastructure & Hardware Mastery: Deep understanding of GPU architectures (NVIDIA RTXAda/Blackwell workstation cards, L40S, A100, H100), VRAM calculation for model weights + KV cache +context windows, multi-GPU tensor parallelism, and Linux server administration.● Inference & Optimization: Hands-on experience with vLLM, TensorRT-LLM, ONNX Runtime, modelquantization techniques, and fine-tuning (LoRA/QLoRA).