Internal AI PlatformLLM ApplicationEngineering Productivity

GenAI Workspace for Engineering Teams

Built as a reusable internal workspace rather than a single-use demo.

AI Application Engineer2025/08 - PresentPixArt

Internal AI Workspace · Routing Flow

Air-Gapped IntranetOn-Prem vLLMShared Engineering Hub
  1. 01

    Engineering Task

    Coding, Docs & Analysis

  2. Core Hub02

    Task Router

    Access Control & Model Pick

  3. 03

    Fast Serving

    Low-Latency Chat & Code

  4. 04

    Deep Reasoning

    Complex Technical Review

  5. 05

    Doc & Vision

    Multimodal Input Utilities

  6. 06

    Team Workspace

    Reusable Daily Patterns

Unifies on-prem model serving, document/vision utilities, and repeatable workflows into one engineering entry point.

On-Prem Serving · Reusable Patterns
Workflow overview. Some names and data are anonymized or recreated for confidentiality.

Before

Engineering users had fragmented access to AI tools, and standalone model endpoints were difficult to use in daily work.

After

Combined internal model access, reusable interaction patterns, and document/image-related utilities into a shared workspace designed around recurring engineering tasks.

Context & Background

  • Engineering teams wanted practical AI support for coding, document reading, and technical analysis.
  • The need was broader than a chatbot: teams needed a place where multiple AI capabilities could be used as part of recurring work.

Constraints

  • Tool choices had to fit internal infrastructure, access controls, and different task depths.
  • A useful solution needed to balance fast responses, stronger reasoning, and document/image-related input handling.

Problem Solved

  • Teams lacked a usable internal way to access multiple AI capabilities inside their daily engineering workflow.
  • Standalone model endpoints did not automatically translate into adoption or reusable workflows.

My Role

  • The problem was treated as a product and workflow design task rather than a pure model deployment task.
  • Access to different model capabilities and repeated-use patterns was shaped around how teams would actually use the workspace.

System Design

  • Built a shared workspace around internal model serving, document/image-related utilities, and task-oriented interaction flows.
  • Organized model access so lightweight tasks and deeper reasoning requests could be handled through clearer workspace patterns.

Implementation Highlights

  • Designed a shared internal workspace rather than a one-off AI tool.
  • Connected model serving, document/image-related utilities, and workflow-oriented access patterns into one shared workspace.
  • Supported both quick-response and deeper reasoning use cases through clearer model access patterns.

Outcome & Impact

  • Created a practical internal GenAI entry point for engineering teams instead of isolated experiments.
  • Made AI support easier to adopt across coding, document review, and technical analysis workflows.

Tech Stack

PythonvLLMModel ServingDocument UtilitiesDocker