Portfolio overview¶
CAE Copilot is the Applied AI / LLM engineering anchor in a dual-portfolio strategy alongside SoccerPredictAI (MLOps platform proof).
Primary document: Problem statement and positioning — why the project exists, five engineering problem areas, design principles, and scope.
Navigation by role¶
| Role | Recommended path | Time |
|---|---|---|
| Recruiter / Hiring Manager | 2-minute path below | 2 min |
| Technical Interviewer | Technical deep-dive below | 15–20 min |
| ML / Applied AI Engineer | Problem statement → data flow → runtime view | 15 min |
| Platform / MLOps Engineer | status.md → deployment | 10 min |
| Engineer reviewing code | quickstart.md → requirements | 20 min |
2-minute path¶
"What is this project and is it real?"
- Problem statement — five problem areas + Rocky scope (skim).
- Status — Stage 2 done; Stage 3 deferred.
- Key facts:
- Stage 2 (shipped): RAG + citations, LangGraph routing, Pydantic tools, eval, HITL orchestrator model
- Stage 1 (done — MVP): Rocky v26r1 PDF corpus, manifest ingest, vector + metadata filters; Unstructured.io planned
- Stage 3 (deferred): async solver orchestration — solver-agent-skill.md
- Demo paths: knowledge/setup/scripting (RAG → Citation → Report) or calculation (Tool → Report)
Live: UI · API (Swagger)
Technical deep-dive (15–20 minutes)¶
"Can this person design agent systems for engineering workflows, not just wrap an LLM?"
| # | Page | Time | Interviewer question |
|---|---|---|---|
| 1 | Problem statement — five problem areas + principles | 5 min | Does the candidate understand domain pain? |
| 2 | Architecture overview | 2 min | Agent graph, not a chatbot? |
| 3 | Data flow + Runtime view | 5 min | RAG path and tool retry loop? |
| 4 | Data-First + Principles | 3 min | RAG vs Tools split? Model-agnostic? |
| 5 | Status + eval metrics | 2 min | What is actually shipped (Stage 2)? |
| 6 | Demo walkthrough | 3 min | Can they demo traceability? |
| 7 | Roadmap — Stage 3 summary only | 2 min | Scope honesty without overpromising? |
Not in deep-dive path: pitch, positioning (agent cheat sheets), deployment (unless infra is the focus).
What this project proves (by competency)¶
| Competency | Evidence |
|---|---|
| Enterprise adoption thinking | Anti-patterns (chaos, AS-IS, black box, lock-in) — problem statement |
| Data engineering | Data-First RAG vs Tools split; vector + metadata filters — data-first.md |
| Agent architecture | LangGraph router, shared CAEState — overview |
| Boundary validation | Pydantic tool schemas, retry loop — runtime view |
| Eval discipline | Golden Q&A, citation hit rate — pdm run eval; status |
| Production patterns | API/UI split, model-agnostic LLM, Helm/k8s — deployment |
| Async job design (planned) | Solver Agent spec, queue pattern — solver-agent-skill.md |
| Scope honesty | Stage 3 deferred; not solver automation as current product — roadmap |
Quick links¶
| Page | Purpose |
|---|---|
| Problem statement | Why · problems · principles · scope |
| Pitch | Short product summary + live links |
| Positioning | One-page cheat sheet for agents |
| Demo walkthrough | 3-minute demo script |
Live surfaces¶
| Surface | URL |
|---|---|
| UI | cae-copilot.dmitryivanov.dev |
| API (Swagger) | api.cae-copilot.dmitryivanov.dev/docs |
| Docs | docs.cae-copilot.dmitryivanov.dev |
Implementation status: status.md.