Implementation Workspace
Video transcript
00:00 — The Implementation Workspace
The Implementation Workspace. Reviewing schematics, simulations and layouts.
00:09 — Welcome to your Implementation Workspace
This is where you bring schematics, SPICE simulations and PCB layouts for an end-to-end review. It is the one place on Creopus where you ask the AI to judge an implementation rather than generate one.
00:27 — Link to a hierarchy node first
Before the AI review can run, the workspace has to be attached to a node in your System Hierarchy. The AI grounds every judgment against that node's requirements, DFMEA, block diagram and BOM — so the same schematic is reviewed differently depending on whether you link it to an automotive subsystem or a medical one.
00:54 — At-a-glance status
Four numbers give you the state of the review — sub-artifacts uploaded, total findings raised, how many are still open, and how many of those are critical. Baseline the workspace whenever you want a frozen reference to come back to.
01:13 — Upload a schematic
Three ways in — paste a netlist, load a file, or attach a PDF or photo of a hand-drawn schematic for the AI to read visually. Around fifty automated and pattern checks run instantly on any netlist it can parse, covering floating nets, missing decoupling, multiple drivers on one net, and more.
01:37 — Upload a SPICE deck
Paste an LTspice, ngspice or QSpice deck. Hygiene rules check transient step size, AC sweep coverage against your RC corners, ground reference and model definitions. Attach waveform screenshots if you want the AI to judge whether the deck actually exercises what you intended.
02:02 — Upload a PCB
Four formats — KiCad, a Gerber zip straight from your fab package, ODB plus plus, or IPC 2581. Automated rules check board outline, trace width, annular ring, drill sizes and copper-to-edge clearance. Pattern checks add decoupling proximity, via-in-pad, and real clearance computed from exact geometry.
02:29 — Your uploaded artifacts live here
Every upload becomes a sub-artifact card, showing its parsed component and net counts and a chip for each attached file. Delete one to re-upload, and its automated findings go with it — the AI findings persist until you re-run the review.
02:51 — AI schematic review
After the instant checks, this runs the grounded AI review. It reads your requirements, DFMEA, block diagram, BOM and any attachments, then judges whether the schematic actually realises the stated intent. It stays disabled until you have linked a node and uploaded a schematic.
03:15 — AI layout review
Same grounding — different surface. The layout reviewer judges the parsed board (or Gerber / ODB++ / IPC-2581) against the same requirements, DFMEA, block diagram and BOM. It comments on routing decisions, EMC tradeoffs, DFM, and whether the layout is consistent with the schematic and sim already in the workspace.
03:47 — AI cross-check
Once you have uploaded at least two of schematic, simulation and layout, this runs a consistency review across them. It surfaces what no single-artifact reviewer can see — a schematic net the simulation never references, or a layout footprint that disagrees with the schematic.
04:09 — Findings — discipline, disposition, discussion
Every review tier funnels into one Findings list, grouped by severity. Each finding carries a discipline badge — Electrical, Layout or Mechanical — and the chips at the top filter by who is responsible. Disposition a finding as accept, reject, defer or note and it collapses, so the open ones stay in focus. You can edit an AI finding without losing its provenance, add one the AI missed, and use the discussion thread inside a finding for the back-and-forth while a fix is in flight.
04:47 — AI follow-on — focused directives
Used the canned review buttons above and want to re-run with a more specific lens? Pick one of the canned directives (e.g., "flag any rail without dedicated decoupling") or write your own — the AI re-runs against the same hierarchy context but biased toward your directive. Existing findings are preserved; new ones inherit the directive's framing in their citations.
05:16 — Multiple workspaces
Each implementation review lives in its own workspace. Click + New workspace to start a fresh one — your past workspaces stay one hamburger-click away (top-left). You can have parallel workspaces for different blocks of the same product, or revisit older reviews to mine their findings.
05:39 — docs.creopus.ai
Full documentation, including every tool, is at docs.creopus.ai
Upload your implementation; get grounded, structured findings.
What it's for
Unlike the generative tools, this one reviews artifacts you upload — schematic netlists, SPICE decks, PCB layouts — through a three-layer pipeline (deterministic rules → curated patterns → an LLM grounded in this node’s requirements, DFMEA, block diagram and BOM) and returns findings grouped by severity, tagged by discipline (EE/LE/ME) and layer (L1/L2/L3).
When to use it
During implementation, to catch issues before fabrication or bring-up.
How it works
- Upload a schematic (netlist/PDF/image), a SPICE deck, or a KiCad PCB / Gerber zip. → It’s parsed (KiCad, Altium, Gerber, ODB++, IPC-2581, SPICE/LTspice/QSPICE…) and Layer-1 rules + Layer-2 patterns run immediately, appending findings.
- Run an AI review — schematic, layout, cross-check, or ✦ Deep AI review. → The Layer-3 LLM review runs async with live streaming, grounded in the node’s requirements, DFMEA, block diagram and BOM; the deep agentic review takes ~30–90s.
- Filter findings by discipline and layer. → Chip filters (All/EE/LE/ME, All/L1/L2/L3 with counts) plus a per-sub-artifact selector and Show/Hide resolved narrow the list; a scoreboard tracks open/accepted/rejected/deferred/noted.
- Disposition a finding and discuss it. → Accept / Reject / Defer / Note / Re-open with an optional reason (audited); each finding has a comment thread, and you can add a manual finding or edit one. Findings link back to Req/DFMEA ids and “📊 Backed by calc”.
Mechanics
- Three-layer review: L1 deterministic rules → L2 patterns → L3 grounded LLM (async, streaming)
- Findings by severity · discipline (EE/LE/ME) · layer (L1/L2/L3)
- Parsers: KiCad · Altium · Gerber · ODB++ · IPC-2581 · SPICE family
- Per-finding disposition + audit + comment threads · manual findings
- Sub-artifact revisions · deep-link + highlight a single finding