Portfolio polish: close production-readiness gaps and strengthen project-specific evidence

[CoreyLeath-code]

Context

A senior AI engineer reviewed the portfolio and recommended converting the remaining caveats into stronger, verifiable evidence.

Goals

• Verify and document the current unit/integration test coverage; establish a realistic enforced threshold in CI.
• Run and document a complete model/system evaluation using reproducible commands and fixed seeds where applicable.
• Confirm security workflows produce visible results; document CodeQL/dependency/container/secret scanning status.
• Measure the built Docker image size and document the exact image/tag and command used.
• Validate the quickstart from a clean clone in a fresh environment.
• Surface the most important verified metrics near the top of the README.
• Add a project-specific Engineering Decisions & Trade-offs section covering the hardest implementation problem, key failure modes, compromises, and next production step.
• Remove or update stale pending claims only after the related evidence exists.
• Correct typos and filename inconsistencies without changing valid public interfaces.

Acceptance criteria

1. A clean-clone quickstart completes using documented commands.
2. CI reports tests and coverage and enforces the documented threshold.
3. Benchmark/evaluation results include environment, methodology, dataset/workload, sample count, and caveats.
4. Security and Docker evidence is linked or summarized in the README.
5. README claims remain honest: measured values are labeled measured, and planned values remain explicitly planned.
6. Changes are submitted through a focused pull request with validation results.

Do not invent metrics or claim production authorization. Use measured repository evidence only.

[CoreyLeath-code]

Portfolio rollout started through repository-scoped PRs because GitHub PRs cannot span repositories.

Opened:

• Disease Prediction Capstone: feat: add opt-in Pinecone retrieval practice Disease_Prediction_Capstone#32
• TransformerForge: feat: add opt-in Pinecone retrieval practice TransformerForge#35
• Facial Emotion Recognition: feat: add opt-in Pinecone retrieval practice Facial-Emotion-Recognition-System#23
• TrojanChat: feat: add opt-in Pinecone retrieval practice TrojanChat#40
• HLM Pinecone implementation: feat: add opt-in Pinecone vector store backend -Hierarchical-Language-Model#7

These changes preserve local/offline backends, add opt-in Pinecone adapters where a vector/RAG boundary already exists, add README badges and hosted benchmark contracts, and keep secrets out of source control. QUANT-LLM-ASSIST is tracked as a follow-up because its indexed FAISS path currently returns 404 through the repository API and needs repository-state reconciliation before patching.