AIGC SaaS and Model-serving Platform

Overview

At SenseTime AIGC, contributed to 0-to-1 delivery of the miaohua generative AI SaaS, owning engineering integration with inference services, microservice decomposition, async jobs, logging, monitoring, and platform operations.

The core value was AI productization: connecting model inference, task state, file storage, post-processing, and user experience into usable, scalable, and operable product systems.

Platform Scope

The miaohua AIGC product work included:

1. AI image generation platform supporting proprietary image models, LoRA training, and third-party open-source model integration.
2. Inference integration around Stable Diffusion / PyTorch, including parameter adaptation, prompt optimization, async jobs, model-file storage, and result post-processing.
3. Microservice boundaries and RPC contracts for identity, model inference, model-file storage, and task delivery.
4. Production engineering across trace logging, monitoring and alerting, containerized deployment, load balancing, and throughput optimization.

Engineering Tradeoffs

The key challenge was bringing generative models into a reliable product system:

• Model call chains needed clear records of parameters, inputs, outputs, latency, and failure reasons.
• Task queues, model services, and file storage had to be separated so long-running inference would not block the web path.
• Prompting, model parameters, post-processing, watermarking, QR overlays, and object storage had to become a reusable pipeline.
• Identity, inference, file, and task-delivery services required explicit RPC contracts, load-balancing policies, scaling boundaries, and fault isolation.

Platform Value

Established model-serving engineering, long-running task orchestration, inference observability, and platform operations, extending these patterns into model-call governance, cost control, and production release workflows for enterprise AI Agent platforms.