jobs in UA Pictures Sdn Bhd

UA Pictures Hiring! Full Time Ai Development Engineer (MCP - Agent Direction) in Perlis - Ricebowl

Ai Development Engineer (MCP - Agent Direction)

Share
Save

Working Location

  • Kangar Perlis Malaysia

Job Description

Responsibilities

About the role

AI Development Engineer (MCP & Agent Direction) is responsible for engineering generative AI capabilities into film and animation production workflows, with focus on MCP toolchain construction, agent workflow orchestration, stylized rendering generation pipelines, and model distillation & compression. The role requires complete practical experience in generative models from training, evaluation to local deployment, capable of completing model compression, inference optimization and effect evaluation loops under local compute conditions, ensuring generation results are controllable, reproducible and accountable.


Key responsibilities

  • MCP server planning, development and governance: encapsulate image generation, video generation, asset retrieval, ComfyUI workflow execution, rendering data export capabilities as standard MCP tool interfaces; implement parameter specifications, permission control, cost marking and audit logs; align with MCP stateless specifications, support capability discovery, self-contained requests and standardized error handling; design MCP service interfaces for DCC integration
  • Agent workflow orchestration and implementation: support multi-step task decomposition and execution in film and animation production scenarios, support long task asynchronous execution, breakpoint continuation, failure retry and manual confirmation; implement Orchestrator-Worker, Fan-out/Fan-in, tool calling, structured output, RAG retrieval enhancement capabilities; establish evaluation metrics for agent success rate, manual intervention rate, retry rate, generation consistency and single-shot cost
  • Stylized rendering generation pipeline construction: manage style model training, evaluation and version management (LoRA / Style Pack / IP-Adapter); apply structural constraints (ControlNet, depth maps, normal maps, edge maps, reference image conditions); maintain character ID consistency and inter-shot style consistency; output layered data for AI neural rendering including depth, normal, Alpha, optical flow, camera trajectory, skeleton metadata
  • Model distillation, compression and inference acceleration: design distillation schemes for image, video and multimodal generation models including teacher/student model training, step distillation, training data construction and effect evaluation; complete quantization and inference acceleration engineering covering FP8, INT4, NVFP4, GGUF, ONNX, TensorRT, vLLM solutions; establish automated evaluation system covering FID/CLIP Score, prompt adherence, temporal stability, artifact rate and single-shot generation cost
  • Local inference service deployment and resource management: containerized deployment, interface encapsulation, multi-tenant isolation, GPU memory quota management, task queue and priority scheduling; establish cost accounting mechanisms for single image, single shot, per-second video generation; monitor service health, queue congestion, failure modes and regeneration trigger conditions
  • Third-party AI model API integration and orchestration: evaluate, integrate and orchestrate multiple third-party AI model APIs (image generation, video generation, 3D model generation, PBR material generation); design API routing strategies, load balancing and degradation schemes ensuring stability and timeliness in multi-model parallel calls
  • DCC software AI client integration: embed AI generation capabilities as plugins in DCC tools (Blender/Maya) enabling artists to call AI capabilities directly in software viewport; design and implement DCC plugin AI function modules including in-viewport AI generation preview, parameter adjustment panels, asynchronous task status monitoring; ensure AI operations integrate seamlessly with artist workflows
  • AI generated content quality evaluation system construction: establish automated quality evaluation pipeline for AI generated content including 3D model topology quality scoring, inter-frame consistency detection for video, material seamlessness checking, image style consistency assessment; establish quality gate standards ensuring AI output meets production archival threshold
  • Agent tool registry and capability discovery mechanism: design and manage MCP tool registry supporting tool version management, capability discovery, dependency declaration and compatibility verification; implement MCP Server capability declaration and auto-discovery protocol enabling AI Agent to dynamically perceive available tools and autonomously orchestrate calling chains
  • Multimodal generation pipeline design: design cross-modal AI generation pipelines (image→video→3D→material) involving format conversion between different models, condition data transmission, style consistency maintenance; design pipeline-level error handling, retry mechanisms and manual review nodes ensuring end-to-end reliability
  • Generation result reproducibility, version management and regression testing: implement model version locking, random seed recording, parameter snapshots, workflow JSON archiving and rendering manifest generation; establish regression testing mechanisms for style packs, distilled models, MCP tools and agent workflows; support version rollback and issue closure when quality, latency or cost baselines regress
  • Technology tracking and production introduction assessment: track technology progress in generative video, stylized rendering, MCP/Agent frameworks, local inference optimization; organize PoC verification and selection evaluation from quality, stability, license compliance and production adaptation dimensions providing actionable introduction recommendations


About you

  • Bachelor's degree or above in Computer Science, Artificial Intelligence, Automation, Computer Graphics or related field
  • 3+ years Python engineering development experience with service-oriented and containerized deployment practice
  • 1+ year generative model engineering experience covering at least two directions among training, evaluation, quantization, distillation or inference optimization
  • Strong English technical literature reading ability and technical documentation writing ability
  • Familiar with mainstream generative model training and inference processes including Stable Diffusion, FLUX / FLUX.2 Klein, Qwen-Image, AnimateDiff, LTX-2, Minimax H3 or equivalent diffusion/flow matching models
  • Deep hands-on experience with ComfyUI and/or Diffusers capable of completing workflow automation execution, custom node integration, model/VAE/LoRA management, batch scheduling and VRAM optimization
  • Master model compression and inference acceleration methods, familiar with at least two of: knowledge distillation, step distillation, FP8/INT4/NVFP4 quantization, LoRA merging, TensorRT, vLLM, GGUF/ONNX export or inference acceleration frameworks
  • Familiar with style fine-tuning training process (LoRA / Style Pack) capable of independently completing data construction, overfitting discrimination, style consistency evaluation and online closure
  • Large language model application development experience, familiar with tool calling, structured output, RAG retrieval enhancement, Function Calling, MCP tool registration and agent evaluation
  • Engineering judgment capability able to reasonably divide boundaries between model capabilities and rule logic, accurately attribute generation, inference and pipeline failures
  • Third-party AI model API integration and orchestration experience with actual engineering experience integrating multiple external AI generation APIs (image/video/3D generation), capable of implementing API routing design, load balancing, degradation strategies and circuit breaker mechanisms
  • DCC software AI client integration experience with actual development experience embedding AI capabilities as plugins in Blender/Maya and other DCC tools, understanding DCC plugin development frameworks, viewport integration mechanisms and artist workflow adaptation
  • AI generated content quality evaluation system construction experience capable of designing automated quality evaluation pipelines covering 3D model quality inspection, video consistency detection, material quality assessment dimensions, establishing quantifiable quality gate standards.

Important Information

Never provide your bank or credit card details when applying for jobs. Do not transfer any money or complete unrelated online surveys. If you see something suspicious, Report this Job ad.

Learn More