AiGenePublicationCommentService implementation

[bobular]

Python source

https://github.com/PubLLicanProject/VPDB_AI_gene_paper_summary/tree/main

[bobular]

Oops the commit I just made should have been "Deliverable 0" - an LLM wouldn't have made that mistake!

---

⏸ Pausing here for review — as the plan directs

"Module builds clean; the three endpoints return 501. Pause here for user review of the shape before any pipeline logic lands."

Deliverable 0 (scaffolding) is complete and verified. Both modules compile clean, every new class is on disk and emits bytecode, and the service is registered. No pipeline logic has landed — every method body is either a 501 response (the three endpoints) or throw new UnsupportedOperationException("… deliverable N"), each tagged with the deliverable that will fill it in.

What I'd like you to eyeball before I start Deliverable 1

The shape is in these files (all under services/ai/ + the Model comment/pojo & comment/repo packages). Worth a look:

• AiGenePublicationCommentService — the JAX-RS surface: path /user-comments/ai-gene-publication, extends AbstractUserCommentService (so it inherits fetchUser()/getCommentFactory()).
• JobState / JobStatus — the wire-contract value types (stages enum, terminal-status enum, follower list).
• migration_comment_b22.sql — the DDL going to Mustafa/Steve.

Two decisions I want your call on

1. AiGenePublicationRequest DTO — I added this POST-body POJO; it wasn't in the plan's file table but the contract implies it. OK to keep?

2. JobRegistry as a process-wide singleton — simplest for v1 (one bounded pool + one eviction scheduler per JVM). Fine, or would you rather it hang off the WdkModel/application lifecycle?

Note on deeper verification

Plan verification #2/#3 (501 over HTTP, 404 for unknown id, 401 for guests) need a deployed instance and only become meaningful once Deliverable 1 adds real routing/auth behavior. If you want, I can deploy and smoke-test the 501s now — otherwise I'd fold those into the Deliverable 1 review.

Want me to proceed to Deliverable 1 (sync prelude), or adjust the scaffolding first? Nothing is committed to git yet either — say the word if you'd like a checkpoint commit before moving on.

[bobular]

It wasn't clear how synonyms and the jobId was wired from SyncPrelude to the main job/pipeline. Hopefully it's clearer now with the latest commit.

---

What changed

File	Change
JobSubmission.java (new)	Immutable carrier of the resolved inputs. Built from (request, jobId, synonyms, modelName, promptVersion, optionsJson) — derives sourceKind/pubmedId/uploadedPaperText/etc. from the request so call sites stay small.
JobState.java	Dropped the per-follower Submitter-with-full-request. Now holds one JobSubmission (shared) + List<Long> followerUserIds + a userId→commentId map filled at persist. getJobId() delegates to the submission.
AiGenePublicationPipeline.java	Constructor is now (JobState, WdkModel); reads inputs via _job.getSubmission(). Added the transient stage-output fields (_articleText, _mentionCounts, _summaryJson, _validatedJson, _aiHeadline, _aiContent) so the ①→⑥ threading is visible.
JobRegistry.java	submit(JobSubmission, long userId, Function<JobState,Runnable> pipelineFactory) and attach(String jobId, long userId). The factory resolves the construction order (pipeline needs the JobState the registry creates).

So, to directly answer your question — synonyms and jobId now flow like this:

SyncPrelude.resolveSynonyms()/computeJobId()
   └─► new JobSubmission(request, jobId, synonyms, model, promptVersion, optionsJson)   [immutable]
         └─► JobRegistry.submit(submission, userId, jobState -> new Pipeline(jobState, wdkModel))
               └─► JobState holds submission; Pipeline reads jobState.getSubmission().getSynonyms()
                     └─► stages thread article text / scan counts / summary via instance fields
                           └─► persist loops followerUserIds → createComment, recordComment(userId, commentId)