Claude Code for Chemical Engineers: Plant Troubleshooting with Agentic AI

Comments · 16

• DE

  @deborahdamilola · 1 day ago

  Amazing stuff Scott. Thanks for sharing 🫡

  1

• WI

  @willlt64 · 13 days ago

  Great stuff Scott, very hard to find this type of content publicly available.

  2

• MA

  @MartyDytrych · 2 weeks ago

  Scott - another banger! Love the depth and insight.

  2

• PA

  @parvizisgandarov5628 · 2 weeks ago

  Thanks for detailed explanation.

  1

• TH

  @TheGabyvanegas08 · 2 weeks ago

  Amazing video and explanation!!!

  1

• ED

  @eduardvendrell9136 · 2 weeks ago

  Thanks for sharing!

• BR

  @bryanfiguresitout · 4 days ago

  Really good stuff! Have you thought about using an open source local model such as Qwen to replace the LLM in your architecture? It seems this could ease the fears of corporate since it could be "air gapped" from the outside world.

  1

• CO

  @cormacgarvey3566 · 2 weeks ago (edited)

  Claude is saying there is an MCP server bug when i try to run the MCP server script - pip piulled latest version that has different SDK&apos;s decorator API .....Root cause found and fixed: requirements.txt had mcp&gt;=1.0 with no upper bound, so pip installed the brand-new mcp 2.2.0, which removed the @server.list_tools()/@server.call_tool() decorator API this codebase depends on — that&apos;s what caused CONNECTION_CLOSED. <br> <br>Fixed by: <br> <br>Installing mcp==1.30.0 (last 1.x release) into .venv — confirmed the server now starts without error. <br>Pinning mcp&gt;=1.0,&lt;2.0 in requirements.txt so this doesn&apos;t happen again on a fresh setup.<br><br>On the great video. I would say the biggest issue is the knowledge graph &nbsp;-it is not formally structured as per something like ISA-s95/S88 which shows the relationships between physical equipment. Simiarily for the control system layer, document layer, historican tag layer. The graph is the meta layer where all of these islands of info meet. Also there are no document nodes in the knowledge graph and it looks like entities and edges were not extracted from the documents either. Ideally there would at least be a link to the docs from the knowledge graph. The docs are semantic information but not engineering fact ,unlike the graph (potentially - this is a database design consideration - only put &nbsp;deterministic stuff in the graph) . So a human review entity extraction process can go into the knowledge graph as a fact and not just reside in the doc as less certain evidence. And there were no design documents at all . No URS, no functional spec etc, no requirements spec. How is the plant supposed to work etc. That is the biggest gap.

  2