Engineering AI
The same four stages in every app.
Requirements, a BOM or a board description go in. Type or paste them, or upload documents and spreadsheets (TXT, MD, PDF, DOCX, CSV or XLSX, depending on the app).
The input becomes a structured model: requirements, BOM lines, functional blocks. What is missing is flagged, not guessed.
The AI model writes the engineering content in a fixed format. An answer that does not validate is sent back with the error to be corrected.
Computed results are added (traceability, savings, derating), weak items are corrected or rejected, and everything is written up and ready to export.
One sign-in, one LLM setting, three jobs.
AI Test Creator
Turn requirements into functional and system test cases, with a traceability matrix and a coverage report.
Done in code
16 agents: 15 AI + 1 in code
AI Cost Optimization
Find savings in an electronics bill of materials: duplicates, alternates, obsolete parts and supplier consolidation.
Done in code
11 agents: 9 AI + 2 in code
AI Component Selection
From board requirements to a schematic-ready main active component recommendation, with derating checks and risks.
Done in code
15 agents: 10 AI + 5 in code
A language model is good at judgement and weak at arithmetic. So the figures an engineer will rely on come from deterministic engines you can reproduce, and the model is used for what it is good at.
Voltage margin, current, surge survival after the clamp, dropout and junction temperature are calculated against the worst case. A 25 V part on a 24 V rail is flagged HIGH RISK.
BOM cost maths, Pareto/ABC and duplicate rules run in code. Each BOM line counts its best saving once, so savings are never double counted.
The requirement traceability matrix and the coverage report are computed from the data, then a final pass covers any requirement still without tests.
A recommendation's confidence is the lowest of the model's judgement, the rating checks and the CRITICAL missing requirements. A design with CRITICAL gaps is never reported as High.
Each app has a Knowledge Base for your own material: standards such as ISO 26262, ASPICE and DO-178C, test guidelines, datasheets, application notes, design rules, approved cross-reference and parts lists. The passages most relevant to each requirement, BOM line or board are given to the model with every run.
After you sign in, Settings offers demo mode (offline sample output, no limit) and your own LLM, which you connect once. Testcase Creator Agent, BOM Cost Optimizer Agent and Component Selector Agent all use your choice, and you can change it at any time. The MithraTiny LLM, our own model, is in future development.
With a cloud provider your prompts go to that provider under its terms. With a server you run, they go to that server.
Results show their reasoning. Every recommendation carries a confidence, its risks and the validation it needs; missing requirements are listed, not hidden. Test cases are editable in the browser, BOM recommendations can be approved, rejected or marked implemented, and every app exports its report.
AI output can be wrong. Verify every design, calculation and test against official datasheets, standards and your own testing before you rely on it.
Run the apps on your own infrastructure with local models such as Qwen, DeepSeek or Llama, a private knowledge base, SSO and dedicated support.
Where Engineering Meets Super Intelligence. Free to start: connect your own OpenAI, Gemini or Ollama once in Settings.