<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Rohan Unbeg on Medium]]></title>
        <description><![CDATA[Stories by Rohan Unbeg on Medium]]></description>
        <link>https://medium.com/@rohan_unbeg?source=rss-ef46c971e707------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Yyb_SbzVNj--IbvpSD0nxA.jpeg</url>
            <title>Stories by Rohan Unbeg on Medium</title>
            <link>https://medium.com/@rohan_unbeg?source=rss-ef46c971e707------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 19:47:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rohan_unbeg/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[GSoC 2026 Final Report: Extend Translation Infrastructure to Exploration Metadata and Skills]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-final-report-extend-translation-infrastructure-to-exploration-metadata-and-skills-a364aa94f096?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/a364aa94f096</guid>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[oppia]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Sun, 06 Sep 2026 14:01:22 GMT</pubDate>
            <atom:updated>2026-09-06T14:49:50.246Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/963/0*_XG_qEv4OfsAuqBm.jpeg" /></figure><p><strong><em>Google Summer of Code 2026 with the Oppia Foundation</em></strong></p><p>Oppia is a free online learning platform built for students who do not have access to good teaching, mostly children between 8 and 15 in places like India, Brazil and Nigeria. A lot of what makes it work is that <strong>volunteers translate the lessons into their own languages</strong>. This summer I worked on the part of Oppia that those volunteers use, and on the parts of a lesson they were never able to translate.</p><h3>The problem</h3><p>If you set Oppia to Hindi and open a lesson, the lesson text comes out in Hindi. Then you look at the title of the lesson, and it is in English. You open the short explanation card that appears when you get stuck, and that is in English too. You go to the library to search for the lesson you were just reading, and you have to type its English name to find it.</p><p>The reason was that Oppia’s whole translation system had been built around one thing: the text inside an exploration card. A translator would be shown a list of “opportunities”, and every opportunity was an exploration. The database model was called ExplorationOpportunitySummaryModel, and everything from the contributor dashboard down to the review flow assumed that was the only kind of thing that could be translated.</p><p>So it was not that nobody had got around to translating lesson titles. There was <strong>no way to express “this title needs translating” at all</strong>.</p><h3>What I built</h3><p>The project had two halves.</p><p>The first half was <strong>making the system stop caring what kind of thing it is translating</strong>. I replaced the exploration-only model with a generic one that stores an entity type alongside the entity id, and then worked through every layer that touched it: the storage queries, the domain services that create and update opportunities, the handlers the dashboard calls, the frontend services, and the pinning feature that lets a contributor pin an opportunity to the top of their list. None of this changed anything a user could see. It was <strong>six weeks of work whose entire visible result was that nothing broke</strong>.</p><p>The second half was actually using it. By the end of the summer a translator can now translate:</p><ul><li><strong>a lesson’s title, objective and tags</strong></li><li><strong>a skill’s description and explanation</strong>, which is what a learner reads on a concept card</li></ul><p>On the contributor dashboard there is now a <strong>Content Type filter</strong> next to the language and topic filters, so a translator can choose to work on lessons, or on skills, or see everything together. Once a reviewer accepts one of those translations, a learner browsing in that language sees the translated title and objective on the library page, sees the translated explanation when they open a concept card mid-lesson, and can <strong>search the library by typing the translated title or tag</strong>.</p><p>The last part turned out to matter more than I expected. Being able to read a lesson in your language but only being able to find it by its English name is a strange kind of half-translated.</p><h3>See it working</h3><p><a href="https://drive.google.com/file/d/1588jpemY7n65NGWWdSaSC2uVrrh2l5Vu/view">https://drive.google.com/file/d/1588jpemY7n65NGWWdSaSC2uVrrh2l5Vu/view</a></p><p><em>Walkthrough of the full flow: a contributor filtering the dashboard by content type, translating a lesson title and a skill explanation, a reviewer accepting them, and a learner seeing the translated title, objective and concept card, then searching for the lesson in Hindi.</em></p><h3>The parts that were hard</h3><p><strong>Working through 20 pull requests in a queue.</strong> This was the thing I underestimated most. Oppia has strict review standards and a lot of the work was sequential, so I would open a PR, wait for review, and want to start the next one that depended on it. Early on I stacked branches on top of each other, which meant the GitHub diff for a child PR showed all of its parent’s files as well, and touching a shared file pulled in a whole extra reviewer team who then had to be waited on. I learned to branch new work off develop whenever the files did not genuinely overlap, and to check the diff file by file before asking for review rather than after.</p><p><strong>Migrating real data.</strong> New models are easy. Existing production data is not. I wrote Apache Beam jobs to create the new opportunity records for every exploration and skill already on the site, plus audit jobs that run the same computation with writes disabled so we could compare before committing to anything.</p><p>Testing those on Oppia’s backup server produced the two most educational bugs of the summer. The first was a 500 error the moment my mentor switched the dashboard to Arabic: Expected translation count for language_code ar to be less than or equal to content_count(0), received 1. My job had been counting translation records directly in the database, but some of those records belonged to content that had since been deleted, or to fields that are not translatable at all. Counting them was fine locally where the data was clean and wrong in production where it was not. The fix was to compute the count the same way the live code does, which ignores the orphaned records.</p><p>The second appeared right after: the “My Contributions” tab hung, with a datastore error asking for an index that did not exist. Locally, the development server quietly adds indexes to index.yaml as new queries run, so I had never seen it. Production requires them to be declared in advance. Filtering suggestions by entity type was a new query this project introduced, so the index had simply never existed anywhere except my laptop.</p><p>Both bugs had the same shape, and it is the lesson I will actually carry: <strong>my local environment was lying to me in ways that were invisible until real data hit the code</strong>.</p><p><strong>A report I could not reproduce.</strong> One of the Arabic reviewers, Samar, rejected a translation she had amended and then found the translated text saved anyway. A reject writing to the database would have been a serious bug, so I traced the flow.</p><p>The contributor dashboard does not send a review to the server when you click accept or reject. It holds the action for 30 seconds behind an undo snackbar (COMMIT_TIMEOUT_DURATION in contributions-and-review.component.ts) and only commits when that timer expires or you move on. Because the queue runs behind the modal, the toast you are looking at does not necessarily belong to the card in front of you.</p><p>Samar had accepted an earlier card, content_19, which committed correctly. The validation message she then saw belonged to content_3, reviewed before it. She closed the modal, rejected the current card, and saw the text saved, so it looked like the reject had written it. The accept had already fired in the background, and rejections never touch the translation models at all.</p><p>So the data was right and the ordering the interface implied was wrong, which is arguably worse, because a reviewer cannot tell the difference. We filed a follow-up to look at whether that 30 second delay is worth what it costs.</p><p><strong>Exams.</strong> My university end-semester exams landed in the middle of backup server testing. For about two weeks it was reviewing comments in the morning, sitting an exam at 1pm, and getting back online at 4pm to read server logs with Hasitha. I would not recommend planning it that way.</p><h3>Where it stands</h3><p>The core work is merged into Oppia’s develop branch.</p><p>I ended my week 18 blog post by saying fingers crossed for the final testing. Every feature at Oppia goes through a pre-launch review where the product manager uses it the way a real contributor would, on a copy of the production server, and either signs it off or sends it back.</p><p>On September 6th, Aanuoluwapo Adeoti marked the review <strong>Completed, approved for launch</strong>, and wrote:</p><blockquote><em>“I played around with the feature, and do not have any issues to report whatsoever. It looks good to me. Thanks for working on it, Rohan!”</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AXMkmhD5OGaF1tZ-GDwSEw.png" /></figure><p>That was the moment the project stopped being a pile of pull requests and became something that is actually going to ship. <strong>I cannot wait to see it go live</strong> and to watch the first translated lesson titles show up on the real site.</p><p>Three things are still open: the acceptance tests, a set of changes from Sean Lip’s feedback on the demo, and the developer wiki page. After those merge and the feature goes to production, one cleanup PR removes the old exploration-only models and the feature flag.</p><h3>What comes next</h3><p>The most useful conversation of my whole summer happened in the last week of it, and it was one question from Sean Lip: after your project, is there anything on the site that a non-English speaker would expect to be translated and still is not?</p><p>I had to go and work out the honest answer, and it was longer than I wanted it to be. <strong>Study guides, topic and subtopic names and descriptions, story and chapter titles, and classroom names and descriptions</strong> are all still English only. Practice questions are too, and I had that one wrong at first and had to correct myself in the thread, which was not fun but was the right thing to do.</p><p>Sean then asked how long the rest would take. That is a much easier question to answer now than it would have been in May, because the mechanism already exists and I had just been through it twice. Study guides are the cheapest, since their sections already carry stable content ids. The others need those ids added first, which means a schema change and a data migration each. Roughly 12 to 14 weeks for all four, against 6 to 8 for study guides on their own.</p><p>That is now a follow-up project on Oppia’s CORE team backlog, scoped so the site gets to the point where everything a learner reads is translatable. I also wrote a developer wiki page walking through every step of adding a new translatable entity type, so whoever picks it up does not have to reverse engineer it from twenty of my pull requests.</p><p>That feels like the right way for the project to end. Not with the feature finished, but with the next four made straightforward.</p><h3>What I learned</h3><p>I came in thinking the hard part of a big feature would be the design. <strong>It was not.</strong> The design was mostly settled by week three. The hard part was everything around it: keeping twenty pull requests reviewable, writing migrations that survive contact with data nobody has looked at in years, and explaining a technical decision clearly enough that a reviewer who has not been living in the code can agree or disagree with it. I got a lot of practice at the last one, mostly by getting it wrong first.</p><p>The other thing I learned is <strong>how much of open source is patience</strong>. Some PRs sat for three weeks. That is not a problem to be solved, it is what it costs to have code reviewed properly by people who did not write it.</p><h3>Thanks</h3><p>To Hasitha Kaushan, my mentor, for reviewing an unreasonable number of pull requests and for sitting on the backup server with me at odd hours. To Chris for keeping me on schedule. To Sean Lip for asking the questions that made the scope right. To Aanuoluwapo Adeoti for testing and approving the launch. And to Kevin, Samar, Nitish and everyone else in the Oppia community who reviewed, tested and put up with my pings.</p><h3>Pull requests</h3><p>Tracking issue: <a href="https://github.com/oppia/oppia/issues/24933">#24933</a></p><p><strong>Milestone 1, generic translation infrastructure</strong></p><ul><li><a href="https://github.com/oppia/oppia/pull/26083">#26083</a> Storage queries for the generic opportunity model</li><li><a href="https://github.com/oppia/oppia/pull/26104">#26104</a> Domain logic for generic and pinned opportunities</li><li><a href="https://github.com/oppia/oppia/pull/26202">#26202</a> Generalized opportunity services</li><li><a href="https://github.com/oppia/oppia/pull/26277">#26277</a> Request handlers for the new model</li><li><a href="https://github.com/oppia/oppia/pull/26327">#26327</a> Resolve suggestion target ids generically</li><li><a href="https://github.com/oppia/oppia/pull/26337">#26337</a> Migrate the frontend to the new architecture</li><li><a href="https://github.com/oppia/oppia/pull/26360">#26360</a> Multi-entity support for opportunity pinning</li><li><a href="https://github.com/oppia/oppia/pull/26187">#26187</a> Beam jobs to backfill exploration opportunities</li><li><a href="https://github.com/oppia/oppia/pull/26392">#26392</a> Beam jobs to backfill pinned opportunities</li><li><a href="https://github.com/oppia/oppia/pull/26458">#26458</a> Gate the acceptance tests behind the feature flag</li></ul><p><strong>Milestone 2, metadata and skill translation</strong></p><ul><li><a href="https://github.com/oppia/oppia/pull/26741">#26741</a> Add the metadata content type</li><li><a href="https://github.com/oppia/oppia/pull/26774">#26774</a> Make skills translatable</li><li><a href="https://github.com/oppia/oppia/pull/26931">#26931</a> Skill opportunity lifecycle</li><li><a href="https://github.com/oppia/oppia/pull/26995">#26995</a> Controllers for skill translations</li><li><a href="https://github.com/oppia/oppia/pull/26997">#26997</a> The Content Type filter on the dashboard</li><li><a href="https://github.com/oppia/oppia/pull/27055">#27055</a> Beam jobs to backfill skill opportunities</li><li><a href="https://github.com/oppia/oppia/pull/26961">#26961</a> Show translated lesson titles and objectives</li><li><a href="https://github.com/oppia/oppia/pull/27083">#27083</a> Show translated concept cards</li><li><a href="https://github.com/oppia/oppia/pull/27111">#27111</a> Search the library in the translated language</li></ul><p><strong>Still in review</strong></p><ul><li><a href="https://github.com/oppia/oppia/pull/27135">#27135</a> Acceptance tests for the new translation flows</li><li><a href="https://github.com/oppia/oppia/pull/27208">#27208</a> Changes from Sean’s feedback on the demo</li><li><a href="https://github.com/oppia/oppia-web-developer-docs/pull/540">oppia-web-developer-docs#540</a> Developer wiki page on adding a new translatable entity type</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a364aa94f096" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Weeks 17 & 18: Merging Milestone 2, Backup Server Testing, and Nearing the Finish Line]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-weeks-17-18-merging-milestone-2-backup-server-testing-and-nearing-the-finish-line-70619f4fc6b4?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/70619f4fc6b4</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[google-summer-of-code]]></category>
            <category><![CDATA[gsoc]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Tue, 25 Aug 2026 18:28:18 GMT</pubDate>
            <atom:updated>2026-08-25T18:28:18.990Z</atom:updated>
            <content:encoded><![CDATA[<p>If you had told me a month ago that I’d be debugging missing Datastore composite indexes on the backup server at 4:00 PM right after walking out of a college exam hall at 2:00 PM, all while explaining to maintainors why a translation rejection didn’t actually write bad data to the database, I would have laughed. But that has been the exact workflow of these final GSoC weeks.</p><p>We are officially at the finish line. Almost all the major Milestone 2 PRs have landed on develop. The backup server is green, auto-merge is active, and after weeks of managing stacked branch dependencies, I can finally see the finish line.</p><h3>Balancing Exams and Reviews</h3><p>One of the biggest personal challenges during this period was managing time. During a critical deployment phase where my mentor, Hasitha, was testing the backfill jobs on the backup server, I had to balance final reviews and updates with my university exams. It was a tight schedule: reviewing PR comments in the morning, taking an exam at 1:00 PM, and logging back on at 4:00 PM to analyze traceback logs, locate database count mismatches, and push index fixes. It was stressful, but seeing the PRs finally land made it worth it.</p><h3>The Milestone 2 Merges</h3><p>Almost all of the core Milestone 2 feature branches have officially landed on develop:</p><ul><li><a href="https://github.com/oppia/oppia/pull/26997"><strong>PR #26997 (PR 2.5)</strong></a><strong>:</strong> Adds the Content Type dropdown component to the dashboard.</li><li><a href="https://github.com/oppia/oppia/pull/26961"><strong>PR #26961 (PR 2.7)</strong></a><strong>:</strong> Displays translated exploration metadata.</li><li><a href="https://github.com/oppia/oppia/pull/27083"><strong>PR #27083 (PR 2.8)</strong></a><strong>:</strong> Displays translated concept cards.</li><li><a href="https://github.com/oppia/oppia/pull/27111"><strong>PR #27111 (PR 2.9)</strong></a><strong>:</strong> Searches lessons by translated language titles and tags.</li></ul><p>With these merged, the codebase has officially adopted the Milestone 2 translation architecture.</p><h3>The “Rejection Saves Translation” Scare</h3><p>Right in the middle of this, we hit a scare on the production server.</p><p>Arabic translation reviewers were blocked on a few lessons with the error “already translated and up-to-date” when trying to accept duplicate suggestions (<a href="https://github.com/oppia/oppia/issues/20961">#20961</a>). During testing, Samar (one of the Arabic reviewers) rejected a suggestion on a card she amended, but noticed the translation was still saved. This created a brief concern that rejections were writing data to the database.</p><p>I investigated the codebase and resolved the confusion:</p><ol><li><strong>The 30-Second Undo Delay:</strong> Oppia’s dashboard holds Accept and Reject actions for 30 seconds to allow reviewers to undo their actions. The loading spinner Samar saw was the timer, not a hung request.</li><li><strong>Out-of-Sync Error Messages:</strong> Samar had actually accepted the last card (content_19), which had no previous Arabic translation and saved correctly. However, because of the 30-second queue, the validation error she saw belonged to the previous card (content_3) she reviewed.</li><li>Closing the modal and rejecting the card afterwards made it look like the rejection saved the data. In reality, the Accept had already fired in the background. Rejections never touch the translation models.</li></ol><p>To prevent this UX confusion, we discussed removing or redesigning the 30-second delay in a follow-up issue.</p><h3>Backup Server Testing &amp; Datastore Indexes</h3><p>With the pipeline cleared, we ran our Apache Beam migration jobs (AuditBackfillSkillOpportunityModelJob and BackfillSkillOpportunityModelJob) on the backup server to populate the new models. This revealed two production-specific issues:</p><h4>The Arabic Translation Count Validation Crash</h4><p>When Hasitha switched the dashboard to Arabic, the server returned a 500 error: Expected translation count for language_code ar to be less than or equal to content_count(0), received 1</p><p>Our Beam job was counting entries in the EntityTranslationsModel directly. However, legacy data contained translation records for content that was later deleted, or fields that aren’t translatable (like skill rubrics). On non-translatable entities, content_count is 0, but the legacy translation records remained, causing the validation to fail on read.</p><p>I updated the backfill job to calculate counts using entity.get_translation_count(), which matches the live runtime logic and correctly ignores orphaned translation records.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PyM0SuYevcDU2lV9" /></figure><h4>The Missing index.yaml Composite Index</h4><p>Right after the count fix, the My Contributions tab started hanging. DevTools showed a 500 error originating from a Datastore index error: 400 no matching index found. recommended index is: - kind: GeneralSuggestionModel properties: - name: target_type - name: created_on direction: desc</p><p>Locally, the dev_appserver emulator automatically appends indexes to index.yaml on the fly as queries run. But production App Engine requires composite indexes to be defined in index.yaml beforehand, or the query fails. Since filtering submitted suggestions by target_type (separating lessons and skills) is a new feature in Milestone 2, the index didn&#39;t exist in production.</p><p>I added the missing composite index to index.yaml, pushed the commit, got it approved by Kevin, and Hasitha successfully redeployed. The dashboard now loads instantly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LBAur62QjRjpdM8M" /></figure><h3>PM Demo and UX Updates</h3><p>Preparing the PM demo was a major milestone. Hasitha and I ran the demo for Aanu (UX Lead) and received valuable feedback:</p><ul><li><strong>Lesson-Specific Concept Cards:</strong> We updated the concept card modal language to prioritize the lesson player’s active study language rather than defaulting to the global site language.</li><li><strong>Manual Language Selection:</strong> We created a tracking issue (<a href="https://github.com/oppia/oppia/issues/27165">#27165</a>) to add a manual language selector directly inside the Concept Card modal as a post-launch enhancement.</li><li><strong>Tag Search:</strong> We got tag search working for this demo.</li></ul><p>I recorded a walkthrough of the entire flow (demoforseanlip.mp4) and shared it with Sean Lip to get final sign-off.</p><p><a href="https://drive.google.com/file/d/1588jpemY7n65NGWWdSaSC2uVrrh2l5Vu/view?usp=drive_web">Watch the Milestone 2 Demo Video</a></p><h3>Next Steps</h3><p>With <a href="https://github.com/oppia/oppia/pull/27055">PR #27055 (PR 2.6)</a> fully approved and in the merge queue, we are almost done. The only remaining pieces are:</p><ul><li><a href="https://github.com/oppia/oppia/pull/27135"><strong>PR #27135 (PR 2.11)</strong></a><strong>:</strong> Fixing minor flakiness and timeout issues in our new acceptance tests.</li><li><a href="https://github.com/oppia/oppia-web-developer-docs/pull/540"><strong>PR #540 (PR 2.12)</strong></a><strong>:</strong> Merging the developer wiki documentation.</li></ul><p>Fingers crossed for the final evaluation!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=70619f4fc6b4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Week 16: Debugging a Click That Never Landed, and Working Around a Full Review Queue]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-week-16-debugging-a-click-that-never-landed-and-working-around-a-full-review-queue-15a47580f8c6?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/15a47580f8c6</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Mon, 10 Aug 2026 20:11:03 GMT</pubDate>
            <atom:updated>2026-08-10T20:11:03.829Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week the skill opportunity lifecycle landed and the downstream PRs were unstacked and opened. This week was about getting those PRs to green, answering two rounds of review on PR 2.4, and picking up work that did not depend on anything already in review.</p><h3>Goals for this Week</h3><ul><li>Address the remaining review comments on PR 2.4.</li><li>Get PR 2.5 passing CI, including the acceptance tests that had been failing for several runs.</li><li>Open the next PR without extending the existing chain of dependencies.</li><li>Make progress on deliverables that need no review to start.</li></ul><h3>What I Accomplished</h3><p><strong>PR 2.5 is now fully green.</strong> Three separate problems were tangled together.</p><p>The first was a real bug. The review modal closed after accepting one suggestion. The dashboard’s active entity type defaults to &#39;all&#39;, and the code branched on targetType !== &#39;exploration&#39;, which is true for &#39;all&#39;, so exploration reviews were taking an untested path. Testing for skills explicitly fixed it, and the reviewer suite went from one passing test to five.</p><p>The second was my own mistake. I replaced two screenshot baselines to match the new UI and broke the older acceptance specs. The old spec and the new one were resolving to the same baseline file, one using its own directory and the other using path.join(__dirname, &#39;..&#39;). They render differently, so each run overwrote the other&#39;s expected image. I reverted the shared baselines and gave the new specs their own folder.</p><p>The third is described below.</p><p><strong>PR 2.4 was mostly answered by deleting code.</strong> Hasitha asked why I had extended a helper with a strict parameter and three @overload declarations. He was right to ask. All of it existed for one call site, and the original code handled the same case by letting the fetcher raise. Reverting it made the file identical to develop, so it dropped out of the diff entirely.</p><p>Something similar applied to a validation branch that was kept alive only by a test fabricating a model with an ID shape the real model cannot have. Both were removed.</p><p>He also found a real gap. I was validating a translation suggestion’s state name but not its content ID, so a suggestion with a correct state name and an invalid content ID passed validation. That is now checked for every entity type, with tests.</p><p><strong>PR 2.8 </strong><a href="https://github.com/oppia/oppia/pull/27083#js-repo-pjax-container"><strong>#</strong>27083</a><strong> was opened off develop rather than stacked.</strong> The plan lists PR 2.5 as a prerequisite, so my first instinct was to wait. Checking the actual file overlap showed there was none: 2.8 touches the concept card handler and component, which no open PR goes near. The dependency was about demo ordering, not code. So it branched off develop and blocks on nothing.</p><p><strong>Drafted the wiki page</strong> on adding a new translatable entity type, which has no code prerequisites and was the obvious thing to write while blocked.</p><h3>What I Learned</h3><p>One acceptance test kept failing on mobile only, on the same step, with the same symptom: the translate button was clicked, nothing errored, and the modal never opened. I added a retry and it failed three times identically, which ruled out flakiness. The useful clue was in the server log. In the thirty seconds after the click there were zero HTTP requests, so the click was not slow, it was never reaching the handler.</p><p>The cause was in Puppeteer’s source. ElementHandle.click() calls _scrollIntoViewIfNeeded() before dispatching, which re-scrolls the element to the centre of the viewport. The framework&#39;s own clickability check runs before that scroll, so its result is stale by the time the event fires. On a 375 by 667 viewport the centre sits under the sticky dashboard header, so the click landed on the header. Desktop passed because at 1920 by 1080 nothing needed scrolling, and only the second item in the list failed because the first was already at the top. Dispatching the click on the element instead of by coordinates fixed it.</p><p>The general lesson is that a check like “this element is clickable” is only true at the moment it runs. Anything that moves the page between the check and the action invalidates it.</p><p>Writing the wiki page taught me something different. This codebase has a rule that translation counts must be recomputed from stored translations rather than incremented, because incrementing drifts, and that drift caused a bug that went unnoticed for over two years. The code does recompute and there is a comment at the call site, but nothing states the rule itself or explains why it matters for the next person adding an entity type. Code can show what it does. Documentation has to explain what to keep doing.</p><h3>What I Struggled With</h3><p>The real difficulty this week was that most of my work was finished and waiting.</p><p>By midweek I had four PRs open with nothing pending from my side, against a schedule that had them merging between late July and early August. Reviewing a project this size takes real time, and my mentor has plenty of work beyond my PRs, so the queue is not a complaint. It does mean the critical path moved from my code to the review cycle, which I cannot speed up by working more.</p><p>What I could control was not being idle and not adding work that was itself blocked. That changed how I chose what to do next. Rather than building on the chain, I checked whether the next PR actually needed the code in review, found it did not, and branched off develop. Rather than waiting to document until things merged, I wrote the wiki page.</p><p>Stacking has a compounding cost I understand better now. PR 2.5 shows seventy changed files on GitHub, but only about a third is its feature. Another third is screenshot baselines, and the rest is PR 2.4 showing through, since a stacked PR is diffed against develop and includes everything below it that has not merged. It makes the PR look much larger than it is, which does not help a reviewer who is short on time, and every revision to 2.4 causes churn in 2.5. Not adding a third level was the most useful decision I made this week.</p><h3>Next Steps</h3><ul><li>Share the wiki draft for review before publishing.</li><li>Start PR 2.9, search with translated tags, also independent of everything in review.</li><li>Agree expectations on the cleanup PR. It depends on every other PR merging and on the feature flag being enabled in production first, so it is worth settling now whether it sits inside this milestone or after it.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15a47580f8c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Weeks 14 & 15: Merging Core Lifecycle PRs, Unstacking Branches, and Finalizing Skill…]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-weeks-14-15-merging-core-lifecycle-prs-unstacking-branches-and-finalizing-skill-1ab098547024?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ab098547024</guid>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Thu, 06 Aug 2026 18:55:35 GMT</pubDate>
            <atom:updated>2026-08-06T18:55:35.779Z</atom:updated>
            <content:encoded><![CDATA[<h3>GSoC 2026 Weeks 14 &amp; 15: Merging Core Lifecycle PRs, Unstacking Branches, and Finalizing Skill Opportunity Backfills</h3><p>These two weeks marked a major turning point for Milestone 2. With foundational PRs landing on develop, the focus shifted to merging the Skill Opportunity Lifecycle (<a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a>), unstacking dependent downstream PRs, and preparing the complete skill translation pipeline for review.</p><h3>Goals for these Weeks</h3><ul><li>Get PR 2.1 and PR 2.2 merged into develop.</li><li>Finalize PR 2.3 (Skill Opportunity Lifecycle), address review feedback, and get it merged.</li><li>Clean up and unstack PR 2.4, PR 2.5, PR 2.6, and PR 2.7 on top of fresh develop.</li><li>Implement and test the Apache Beam backfill job for skill translation opportunities (PR 2.6).</li><li>Address UI feedback for the Entity Type Selector component (PR 2.5).</li></ul><h3>What I Accomplished</h3><p><strong>Merged PR 2.1 and PR 2.2 into develop.</strong> Both <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> (Metadata Translations) and <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> (Making Skill domain object translatable) successfully completed the merge queue and landed on develop, clearing the path for the rest of Milestone 2.</p><p><strong>Merged PR 2.3 (Skill Opportunity Lifecycle).</strong> Addressed all review comments from Hasitha and Nikhil on <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a>. This included refining topic deletion cleanup rules in domain services, handling subtopic skills properly, and fixing CI coverage failures. To resolve an automated reviewer block where unrelated codeowner groups (like Android compatibility reviewers) were pinged due to transient config file touches, I reverted those minor files back to develop. Kevin helped unassign the extra reviewer group on GitHub, officially placing PR 2.3 into the merge queue and landing it on develop.</p><p><strong>Unstacked and cleaned PR 2.4.</strong> With PR 2.3 in develop, updated <a href="https://github.com/oppia/oppia/pull/26995">PR #26995</a> (updating controllers for skill translations). Reverted accidental formatting changes across unrelated files to ensure the diff strictly contains the 6 target controller and domain files.</p><p><strong>Refined PR 2.5 (Entity Type Selector component).</strong> Cleaned up formatting diffs on <a href="https://github.com/oppia/oppia/pull/26997">PR #26997</a> to focus strictly on the EntityTypeSelectorComponent and Contributor Dashboard services. Received feedback from Hasitha regarding UI layout and label clarity, and started aligning the three dropdown selectors into a single horizontal row.</p><p><strong>Recreated and opened PR 2.6 on a clean branch.</strong> Built <a href="https://github.com/oppia/oppia/pull/27055">PR #27055</a> directly off fresh develop. This PR implements BackfillSkillOpportunityModelJob and AuditBackfillSkillOpportunityModelJob to populate missing skill opportunity entries in Datastore. Added unit tests for both Beam jobs, with all 18 unit tests passing 100%.</p><p><strong>Finalized PR 2.7 for review.</strong> Updated <a href="https://github.com/oppia/oppia/pull/26961">PR #26961</a> to display translated exploration metadata in learner-facing cards, verified end-to-end Hindi translation rendering, and attached visual proof.</p><h3>What I Learned</h3><p>The review discussion on <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> highlighted the importance of keeping service boundaries and PR scopes clean. Rather than adding thin wrapper functions for skill-specific logic, the right pattern was to call shared service methods directly, the same way exploration opportunities do it.</p><p>Crucially, I also learned that accidentally modifying files outside your PR scope (like parameter registry or storage test files) can trigger automated codeowner pings for unrelated teams (such as Android compatibility reviewers), blocking an otherwise fully approved PR from entering the merge queue. Keeping PR diffs strictly limited to target files prevents these blocks.</p><p>The cleanest approach I learned was to maintain clean feature patches and recreate downstream branches directly off fresh develop once their parent PR merges. This keeps PR diffs strictly limited to their feature files, making reviews faster and preventing CI noise.</p><p>On the technical side, working on the Apache Beam backfill job (BackfillSkillOpportunityModelJob) gave me deeper insight into batch data processing in Oppia. Ensuring the job properly links skills to their respective topics via create_translation_opportunity() while handling missing or deleted topics gracefully was key to writing reliable migration pipelines.</p><h3>What I Struggled With</h3><p>The biggest challenge during this period was managing branch dependencies while waiting for review passes. When multiple PRs depend on each other sequentially, review turnarounds on base PRs can delay downstream work. Re-creating clean branches off fresh develop resolved the diff bloat, but required extra care to ensure no feature changes were lost during branch resets.</p><p>Additionally, handling UI feedback on component layouts required balancing design consistency with strict Oppia style guides. Aligning dropdown selectors neatly while preserving mobile responsiveness takes careful template structuring.</p><h3>Next Steps</h3><ul><li>Get PR 2.4 (<a href="https://github.com/oppia/oppia/pull/26995">#26995</a>) approved and merged into develop.</li><li>Complete the single-row layout update for PR 2.5 (<a href="https://github.com/oppia/oppia/pull/26997">#26997</a>) and request review.</li><li>Support review iterations on PR 2.6 (<a href="https://github.com/oppia/oppia/pull/27055">#27055</a>) and PR 2.7 (<a href="https://github.com/oppia/oppia/pull/26961">#26961</a>).</li></ul><h3>PR Link Reference</h3><ul><li><strong>PR 2.1 (Metadata translations)</strong>: <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> — Merged</li><li><strong>PR 2.2 (Make Skill translatable)</strong>: <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> — Merged</li><li><strong>PR 2.3 (Skill opportunity lifecycle)</strong>: <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> — Merged</li><li><strong>PR 2.4 (Controllers for skill translations)</strong>: <a href="https://github.com/oppia/oppia/pull/26995">PR #26995</a> — Under Review</li><li><strong>PR 2.5 (Entity-type selector UI)</strong>: <a href="https://github.com/oppia/oppia/pull/26997">PR #26997</a> — Under Review</li><li><strong>PR 2.6 (Beam backfill job)</strong>: <a href="https://github.com/oppia/oppia/pull/27055">PR #27055</a> — Under Review</li><li><strong>PR 2.7 (Translated exp metadata UI)</strong>: <a href="https://github.com/oppia/oppia/pull/26961">PR #26961</a> — Under Review</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ab098547024" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Weeks 12 & 13: PR Reviews, Schedule Slip, and Opening PRs in Parallel]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-weeks-12-13-pr-reviews-schedule-slip-and-opening-prs-in-parallel-feb9d8ab2020?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/feb9d8ab2020</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[gsoc]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Wed, 29 Jul 2026 16:46:51 GMT</pubDate>
            <atom:updated>2026-07-29T16:46:51.877Z</atom:updated>
            <content:encoded><![CDATA[<p>These two weeks were all about pushing existing PRs forward and staying on track despite review delays. The main work involved addressing review feedback on <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a>, <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a>, and <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a>, running acceptance test stress tests, and opening <a href="https://github.com/oppia/oppia/tree/gsoc-task-2.4">PR 2.4</a> and <a href="https://github.com/oppia/oppia/tree/gsoc-task-2.7">PR 2.7</a> in parallel to avoid getting blocked while waiting for merges.</p><h3>Goals for these Weeks</h3><ul><li>Get PR 2.1 and PR 2.2 approved and into the merge queue.</li><li>Implement and open PR 2.3 (skill opportunity lifecycle in domain services).</li><li>Address Hasitha’s review comments on PR 2.3.</li><li>Open PR 2.4 and PR 2.7 in parallel to stay on track for Milestone 2.</li><li>Submit a Schedule Slip Report to org admins.</li></ul><h3>What I Accomplished</h3><p><strong>PR 2.1 and PR 2.2 fully approved and in merge queue.</strong> After addressing all review comments and running the acceptance test stress runs, both <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> and <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> received full approval. For PR 2.1, Mohak had requested a 200-run stress test citing a separate CI flake in develop. After clarifying that the referenced issue was an unrelated flake in the story editor on develop and that our PR had already passed 50 clean runs fixing the V2 modal auto-close bug, both PRs were approved and added to the merge queue.</p><p><strong>Opened PR 2.3.</strong> Implemented the full skill opportunity lifecycle in domain services (<a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a>). create_translation_opportunity() now supports skill entities when the V2 feature flag is enabled. Lifecycle hooks were integrated in topic_services.py and skill_services.py to create, update, and delete skill translation opportunities when skills are assigned to topics, updated, or deleted. Also added auto_reject_translation_suggestions_for_skill_ids() in suggestion_services.py and full unit test coverage across all four test files.</p><p><strong>Addressed all PR 2.3 review comments from Hasitha.</strong> Removed the add_new_skill_opportunities() and update_skill_opportunity_on_skill_change() wrapper functions and now calling create_translation_opportunity() and compute_translation_opportunity_models_with_updated_entity() directly, consistent with how exploration opportunities work. Added remove_topic_from_translation_opportunities() to handle topic unpublishing and deletion — this removes the topic_id from TranslationOpportunityModel.topic_ids, deletes the model and auto-rejects suggestions if topic_ids becomes empty, or just updates the list if the skill belongs to other topics. Also switched to topic.get_all_skill_ids() so skills inside subtopics are included. Fixed a CI coverage failure caused by missed lines in skill_services.py.</p><p><strong>Opened PR 2.4</strong> (branched off 2.3). Updates SuggestionHandler in suggestion.py to accept target_type=skill for SUGGESTION_TYPE_TRANSLATE_CONTENT, with full unit test coverage.</p><p><strong>Opened PR 2.7</strong> (branched off 2.1). Updates summary_services.get_displayable_exp_summary_dicts() to batch-fetch entity translations and override title, objective, category, and tags when display_in_language_code is passed.</p><p><strong>Submitted Schedule Slip Report</strong> to org admins and shared a catch-up plan with Hasitha, targeting all Milestone 2 PRs by the August 19 evaluation deadline.</p><h3>What I Learned</h3><p>The review discussion on <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> highlighted the importance of keeping service boundaries clean. Rather than adding thin wrapper functions for skill-specific logic, the right pattern was to call shared service methods directly, the same way exploration opportunities do it. This also avoids naming confusion since “Skill Opportunities” already refers to question-related SkillOpportunityModels in the codebase.</p><p>Working on the skill opportunity lifecycle also showed me how interconnected topic, skill, and opportunity services are. A single action like removing a skill from a topic has cascading effects across multiple models. Designing the cleanup logic carefully — removing topic_id from TranslationOpportunityModel.topic_ids rather than blindly deleting the model — was important to handle skills shared across multiple topics.</p><p>Opening PR 2.4 and PR 2.7 in parallel while <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> was still in review gave me a practical lesson in dependency management: branching off an unmerged PR works fine as long as you are careful about what logic you depend on from the parent branch.</p><h3>What I Struggled With</h3><p>Review turnarounds have been the main bottleneck. <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> and <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> have been fully approved and sitting in the merge queue for several days. Since PR 2.4 and PR 2.5 depend on these merging first, this creates downstream delays. The workaround is to branch off and implement in parallel, but the code cannot be officially reviewed or merged until the parent branches land.</p><p>Syncs were also harder to schedule this period due to both Hasitha and me missing a few meetings, which slowed down real-time alignment. We managed to stay unblocked by communicating asynchronously through PR comments and the team chat.</p><h3>Next Steps</h3><ul><li>Get <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> reviewed and merged.</li><li>Finalize and open PR 2.4 and PR 2.7 once parent branches land.</li><li>Start work on PR 2.5 (entity-type-selector frontend component).</li></ul><h3>PR Link Reference</h3><ul><li><strong>PR 2.1 (Metadata translations)</strong>: <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> — Approved / Merge Queue</li><li><strong>PR 2.2 (Make Skill translatable)</strong>: <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> — Approved / Merge Queue</li><li><strong>PR 2.3 (Skill opportunity lifecycle)</strong>: <a href="https://github.com/oppia/oppia/pull/26931">PR #26931</a> — Under Review</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=feb9d8ab2020" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Weeks 10 & 11: Merging Milestone 1, Passing Midterms, and Aligning on UX]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-weeks-10-11-merging-milestone-1-passing-midterms-and-aligning-on-ux-b30c6b67ad3f?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/b30c6b67ad3f</guid>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Sat, 18 Jul 2026 16:25:33 GMT</pubDate>
            <atom:updated>2026-07-18T16:25:33.414Z</atom:updated>
            <content:encoded><![CDATA[<p>These past two weeks marked a major transition point for my GSoC project. We officially wrapped up and merged the final remaining pieces of <strong>Milestone 1</strong>, passed the <strong>GSoC Midterm Evaluation</strong>, and immediately pivoted into <strong>Milestone 2</strong> with a focus on refining validation constraints and aligning on user experience designs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sCssczO7k1o3hR2VTatHRg.png" /></figure><h3>Goals for these Weeks</h3><p>My main goals for these weeks were:</p><ul><li>Test, verify, and merge the final <strong>Beam migration job</strong> (<a href="https://github.com/oppia/oppia/pull/26187">PR #26187</a>) on the <strong>GAE backup server</strong>.</li><li>Resolve timing flakes in the <strong>E2E acceptance tests</strong> (<a href="https://github.com/oppia/oppia/pull/26458">PR #26458</a>) and pass the <strong>50-run stress test</strong>.</li><li>Submit the <strong>Midterm Evaluation</strong> and review feedback.</li><li>Address reviewers’ comments on <strong>PR 2.1</strong> (<a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a>) (constants, dynamic validation prompts, and strict type testing).</li><li>Align on the dashboard sorting design for non-state translatable content and open <strong>PR 2.2</strong> (<a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a>).</li><li>Attend the scheduled <strong>Break Week Check-in Meeting</strong> with GSoC org admins.</li></ul><h3>What I Accomplished</h3><p><strong>Milestone 1 Completed and Merged:</strong></p><ul><li><strong>PR 1.9 (Beam Translation Backfill) Merged:</strong> After deploying to the backup server, we verified that the job correctly filters translations corresponding to the active exploration version and successfully cleans up orphaned opportunities. <a href="https://github.com/oppia/oppia/pull/26187">PR #26187</a> has been merged into develop.</li><li><strong>PR 1.11 (Acceptance Tests) Merged:</strong> To resolve the timing flakes, I found that window event listeners (like beforeunload) were leaking between Puppeteer runs in the story editor. I refactored them to use arrow functions and added cleanups in ngOnDestroy(). The E2E tests successfully passed the <strong>50-run stress test</strong> and <a href="https://github.com/oppia/oppia/pull/26458">PR #26458</a> has been merged. All 11 PRs for <strong>Milestone 1</strong> are now successfully integrated!</li></ul><p><strong>Passing the Midterm Evaluation:</strong> I received the official <strong>GSoC Midterm Evaluation</strong> pass notification. I reviewed the valuable feedback from my mentor, Hasitha, focusing on improving my devlog consistency, engaging more with the community, and spending more time on critical design analysis.</p><p><strong>Helping New Contributors (GFI Breakdowns):</strong> As part of my <strong>Milestone 1</strong> community contribution requirements, I researched and wrote 8 detailed technical breakdowns for <strong>Good First Issues (GFIs)</strong> on GitHub. These guides provide step-by-step instructions to help new developers get comfortable contributing to Oppia:</p><ul><li><a href="https://github.com/oppia/oppia/issues/15185">#15185</a></li><li><a href="https://github.com/oppia/oppia/issues/16089">#16089</a></li><li><a href="https://github.com/oppia/oppia/issues/16150">#16150</a></li><li><a href="https://github.com/oppia/oppia/issues/18480">#18480</a></li><li><a href="https://github.com/oppia/oppia/issues/23784">#23784</a></li><li><a href="https://github.com/oppia/oppia/issues/20766">#20766</a></li><li><a href="https://github.com/oppia/oppia/issues/13055">#13055</a></li><li><a href="https://github.com/oppia/oppia/issues/20239">#20239</a></li></ul><p><strong>Learning Code Review (Shadow Reviews):</strong> I completed 3 shadow reviews under Hasitha’s guidance. Since reviewing other developers’ code is a key codeowner skill, Hasitha advised that I start with smaller pull requests to ease into the review workflow. I plan to complete the remaining codeowner reviews in <strong>Milestone 2</strong> as more candidate PRs are opened. You can view the pull requests I shadowed here:</p><ul><li><a href="https://github.com/oppia/oppia/pull/26730">PR #26730</a></li><li><a href="https://github.com/oppia/oppia/pull/26729">PR #26729</a></li><li><a href="https://github.com/oppia/oppia/pull/26060">PR #26060</a></li></ul><p><strong>Break Week Check-in Meeting:</strong> On July 14, I attended the <strong>Break Week Check-in</strong> meeting organized by the GSoC org admins. It was a great opportunity to connect and reflect on the milestone 1.</p><p><strong>Refining Metadata Translations (PR 2.1):</strong> I addressed review comments on <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> by replacing hardcoded length validation limits with constants (feconf.EXPLORATION_TITLE_TRANSLATION_CHAR_LIMIT = 36). I also updated getFormattedContentType() to format prompts dynamically based on content IDs, ensuring that translators see specific prompts (&quot;title&quot;, &quot;objective&quot;, &quot;tags&quot;) instead of a generic &quot;metadata&quot; label.</p><p><strong>Sorting Consensus &amp; In-Code Documentation:</strong> I started an email thread with Sean Lip and Aanu to establish how non-state cards (metadata and skills) should be ordered in the contributor dashboard. Sean approved our proposed layout (sorting and appending metadata at the end of state-specific cards) and deferred to Aanu. To address Brian’s review suggestion, I documented this UX decision inside contribution-and-review.service.ts so future contributors understand why we do not combine and sort the entire list.</p><p><strong>PR 2.2 Opened for Review:</strong> I opened <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> to make the Skill domain object inherit from BaseTranslatableObject, registering skill explanations and misconception feedback under synthetic content IDs (feedback_&lt;misconception_id&gt;).</p><h3>What I Learned</h3><p>This week highlighted the intersection of design, engineering, and mentoring.</p><p>On the engineering side, discussing sorting logic on the email thread with Sean and Aanu taught me that exploration-wide metadata (like Title and Objective) does not map to a playthrough sequence. Appending it separately on the dashboard allows translators to focus on lesson content first, and documenting this UX decision in-code keeps the codebase maintainable for future contributors.</p><p>On the community side, writing GFI breakdowns, performing shadow reviews, and attending the org admin check-in showed me how much effort goes into maintaining a large open-source project. Guiding newer contributors and learning how to critique code constructively under a mentor’s guidance are just as critical as writing the features themselves.</p><h3>What I Struggled With</h3><p>Debugging Puppeteer timing flakes locally requires a lot of patience. Finding event listener leaks that only crash the headless CI container after dozens of runs meant tracing the destruction lifecycle of Angular components line-by-line.</p><h3>Next Steps</h3><p>We are currently awaiting Aanu’s thoughts on the card sorting thread and reviews on <strong>PR 2.2</strong> (<a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a>). Once <strong>PR 2.1</strong> (<a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a>) is merged, we will begin extending the controllers and services for skill translations (Tasks 2.3 and 2.4).</p><h3>PR Link Reference</h3><ul><li><strong>PR 1.9 (Beam Translation Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26187">PR #26187</a> (Merged)</li><li><strong>PR 1.10 (Beam Pinning Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26392">PR #26392</a> (Merged)</li><li><strong>PR 1.11 (Acceptance Tests)</strong>: <a href="https://github.com/oppia/oppia/pull/26458">PR #26458</a> (Merged)</li><li><strong>PR 2.1 (Metadata translations)</strong>: <a href="https://github.com/oppia/oppia/pull/26741">PR #26741</a> (Under Review)</li><li><strong>PR 2.2 (Make Skill translatable)</strong>: <a href="https://github.com/oppia/oppia/pull/26774">PR #26774</a> (Under Review)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b30c6b67ad3f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Weeks 8 & 9: Testing Beam Jobs and Resolving Acceptance Test Flakes]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-weeks-8-9-testing-beam-jobs-and-resolving-acceptance-test-flakes-c98a2da2bd4b?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/c98a2da2bd4b</guid>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Thu, 09 Jul 2026 18:25:03 GMT</pubDate>
            <atom:updated>2026-07-09T18:25:03.400Z</atom:updated>
            <content:encoded><![CDATA[<p>Over the last two weeks, the main focus has been on finalizing the remaining PRs for Milestone 1. This involved extensive testing of the Apache Beam jobs on the GAE backup server, resolving merge conflicts, and addressing flaky Puppeteer acceptance tests. Since most of the Milestone 1 implementation was complete on my side, my mentor Hasitha suggested I begin researching Milestone 2 tasks in parallel.</p><h3>Goals for these Weeks</h3><p>My primary goals were:</p><ul><li>Test and verify the Apache Beam migration jobs (PR 1.9 and PR 1.10) on the backup server and merge them.</li><li>Address the 50-run stress test requirement for the Puppeteer acceptance tests (PR 1.11) and resolve any flakes.</li><li>Keep moving forward with the codeowner review requirement and plan the design for Milestone 2.</li></ul><h3>What I Accomplished</h3><ul><li><strong>PR 1.10 (Beam Pinning Backfill) Merged:</strong> This PR backfills the entity_type field for legacy PinnedOpportunityModels to set them to exploration. It also includes a frontend fix that updates the language selector component to fallback and select the first available language if no preferred language is stored, preventing 400 Bad Request crashes when pinning a lesson. During testing, the PR sat idle while the GAE build size limits were being cleared, causing oppiabot to mark it as stale. After updating the branch with develop, Hasitha ran the job on the backup server and noted that the log output showed PINNED OPPORTUNITY MODELS BACKFILLED SUCCESS: 5. I clarified that because the job queries and updates all models, printing the total count of 5 was expected. To verify the change, Sean Lip ran a direct GQL query in the Google Cloud Datastore Console and confirmed all 5 entities had the property successfully set to exploration. The PR was merged 4 days ago.</li><li><strong>PR 1.9 (Beam Translation Backfill) Merged:</strong> This was the major datastore migration PR. During the initial backup server test run, we noticed a discrepancy where the job counted translations from historical exploration versions. I updated the job to filter and only count translations matching the latest active exploration version (t.entity_version == exp.version). I also added orphan-deletion and cleanup logic to automatically remove stale opportunities when explorations are updated or deleted. The second run on the backup server completed successfully, and the PR was merged yesterday.</li><li><strong>Working to Resolve E2E Flakes on PR 1.11:</strong> To merge the final PR, Jayam requested a 50-run stress test on all four Puppeteer acceptance tests. During the initial runs, a couple of timing flakes occurred, and I have been working to investigate and resolve them:</li><li><strong><em>Story Editor Flake:</em> </strong>Discovered that unbound beforeunload and storage event listeners were leaking between runs. I converted them to arrow functions to maintain lexical context and added cleanup logic in ngOnDestroy(). I also mocked removeEventListener in the unit tests to prevent Karma test crashes.</li><li><strong><em>Custom Page Navigation:</em></strong> I refactored the custom goto page navigation utility in puppeteer-utils.ts to reload the page only when navigating to the identical URL. To support this hash-navigation transition, I added a missing waitForSelector to expectControlsNotAvailable in super-admin.ts.</li><li><strong><em>Status:</em></strong> I have pushed these fixes and restarted the stress test runs to confirm that the changes resolve the flakes.</li><li><strong>Good First Issues (GFIs):</strong> The 8 technical GFI breakdowns I wrote are now live and tagged on GitHub. New contributors have already started claiming them and opening PRs based on the guides.</li></ul><h3>What I Learned</h3><p>This period highlighted the importance of clean setups in E2E testing. Failing to clean up window event listeners in Angular components might not fail during local single runs, but running a stress test 50 times in a headless CI container will eventually cause random timeouts. Ensuring cleanup in ngOnDestroy() is essential for test stability.</p><p>I also gained practical experience with staging and verifying production data backfills. The version mismatch discrepancy on the Beam job demonstrated why backup server testing is crucial before touching production datastores.</p><h3>What I Struggled With</h3><p>Stress-testing E2E tests is time-consuming, as any environment hiccup or timing flake can delay approvals and require a rerun of the entire suite. Additionally, coordinating communication over the last two weeks was slightly challenging due to scheduling conflicts, but we managed to resolve all blockers asynchronously through documentation and comment threads.</p><h3>Next Steps</h3><p>Now that both Beam jobs are merged, the focus is on successfully completing the stress runs, getting Jayam’s final sign-off, and merging PR 1.11. Once Milestone 1 is closed, we will kick off Milestone 2 by extending generic opportunity support to exploration metadata and categories.</p><h3>PR Link Reference</h3><ul><li><strong>PR 1.9 (Beam Translation Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26187">PR #26187</a> (Merged)</li><li><strong>PR 1.10 (Beam Pinning Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26392">PR #26392</a> (Merged)</li><li><strong>PR 1.11 (Acceptance Tests)</strong>: <a href="https://github.com/oppia/oppia/pull/26458">PR #26458</a> (Under Review)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c98a2da2bd4b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Week 7: Finishing Milestone 1 Integration, Live Demos, and Backup Server Hurdles]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-week-7-finishing-milestone-1-integration-live-demos-and-backup-server-hurdles-c8ff8a706651?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/c8ff8a706651</guid>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Fri, 26 Jun 2026 07:22:29 GMT</pubDate>
            <atom:updated>2026-06-26T07:22:29.055Z</atom:updated>
            <content:encoded><![CDATA[<p>This week was busy and exciting. We successfully merged the frontend V2 migration, showcased the workflow live in the PM demo, and started setting up the acceptance tests under the feature flag. With the Milestone 1 evaluation starting on June 29, there was a lot to wrap up.</p><h3>Goals for this Week</h3><p>My main goals for this week were:</p><ul><li>Get PR 1.5 (Suggestion Resolver) and the massive PR 1.6 (Frontend V2 Integration) merged.</li><li>Showcase the live V2 translation suggestion lifecycle in the PM demo meeting.</li><li>Address reviewer feedback on the PR 1.9 Beam job.</li><li>Open PR 1.11 to gate acceptance tests under the feature flag.</li></ul><h3>What I Accomplished</h3><ul><li><strong>PR 1.5 and PR 1.6 Merged</strong>: Both PRs are successfully merged. PR 1.6 was a major milestone, with over 2,000 lines of changes refactoring frontend services (like ContributionOpportunitiesBackendApiService, TranslateTextBackendApiService, and TranslateTextService) to target our V2 endpoints. PR 1.5 updates the backend handlers to resolve target IDs using the new generic models.</li><li><strong>Milestone 1 PM Demo</strong>: On June 20, I showcased the V2 translation suggestion lifecycle live on my local setup in front of the mentors and Org Admin (Hasitha, Chris). During the live run, we noticed a minor bug where the progress bar on the dashboard cards did not update when suggestions were accepted. I fixed this right after the demo by checking for null or undefined explicitly (since 0 was being treated as falsy in the progress check). To show the final working state, I recorded a demo video showing the dashboard progress bar fix and the complete V2 translation player flow, which is linked in <a href="https://github.com/oppia/oppia/pull/26337#issuecomment-4762172188">my PR comment</a>.</li><li><strong>Beam Job Refactoring (PR 1.9)</strong>: Based on feedback from the PM demo, I added an override_metadata_feature_flag argument to get_translatable_contents_collection(). This allows the Beam job to bypass the feature flag gate when computing content and translation counts. I also updated the audit job to compare and aggregate total content and translation counts across models to ensure there is no data drift in production.</li><li><strong>Opened PR 1.11 for Acceptance Tests</strong>: I gated the Puppeteer E2E tests behind the V2 feature flag by promoting it to the TEST stage. I also updated the test files to use the V2 opportunity structures and adjusted target names to use the new topics/story hierarchy, such as using &quot;Fractions&quot; and &quot;States of Matter&quot; instead of legacy mock topics.</li><li><strong>Completed GFI Requirements</strong>: I finished my GSoC Milestone 1 community contribution requirements by writing technical breakdowns for 8 Good First Issues (GFIs) on GitHub.</li></ul><h3>What I Learned</h3><p>This week had some valuable lessons on testing and communication.</p><p>On the technical side, working on E2E acceptance tests showed me how different browser-driven tests are from unit tests. When running Puppeteer tests in the TEST stage under a feature flag, every setup detail, such as assigning the ReleaseCoordinator role to enable the flag and ensuring the curation data aligns with the V2 structure, needs to be exactly correct.</p><p>On the communication side, I got a reminder from my mentor Hasitha about keeping my daily devlogs updated. I had fallen behind on them due to project work. I apologized, realized the gap, and have been posting them daily since then to ensure my progress and blockers are clear to the team.</p><h3>What I Struggled With</h3><p>Debugging Puppeteer tests locally can be difficult since they run in headless mode. When the language selector dropdown did not appear in the exploration player during tests, I had to trace the controller logic step by step. I eventually found a database context bug in story_services.py where programmatically created stories were not registering their linked explorations in the ExplorationContextModel.</p><p>Additionally, we faced some delays in testing the Beam jobs (PR 1.9 and PR 1.10) on the backup server. Since Hasitha handles the server admin work and deployments, we have been working to resolve GAE environment limitations:</p><ul><li><strong>GAE Size Limits</strong>: Deploying to the backup server got blocked because the Release Candidate (RC) build size exceeded the maximum allowed limit in GAE. We are currently waiting for an infrastructure fix to land on develop.</li><li><strong>RC Expiration</strong>: The backup server GAE builds expire every 24 hours. If testing gets delayed, Hasitha has to rebuild the RC, which makes the verification cycle slower and delays our merges.</li><li><strong>Codeowner reviews</strong>: I am also working to complete the codeowner skill requirement (reviewing 4 PRs). Because there are not many community PRs under my mentor Hasitha’s scope, I have only been able to shadow-review 1 PR so far. I hope to find more review opportunities in Milestone 2.</li></ul><h3>Next Steps</h3><p>The immediate goal is to resolve the backup server RC build size issues so we can test and merge PR 1.9 and 1.10, and get PR 1.11 (acceptance tests) approved before the evaluation starts on June 29. After Milestone 1 is merged, we will start Milestone 2, beginning with adding ContentType.METADATA and extracting translatable metadata fields from explorations.</p><h3>PR Link Reference</h3><p>For those interested in following the implementation details and review discussions, you can find the PR links below:</p><ul><li><strong>PR 1.5 (Suggestion Resolver)</strong>: <a href="https://github.com/oppia/oppia/pull/26327">PR #26327</a> (Merged)</li><li><strong>PR 1.6 (Frontend V2 Integration)</strong>: <a href="https://github.com/oppia/oppia/pull/26337">PR #26337</a> (Merged)</li><li><strong>PR 1.9 (Beam Translation Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26187">PR #26187</a> (Under Review)</li><li><strong>PR 1.10 (Beam Pinning Backfill)</strong>: <a href="https://github.com/oppia/oppia/pull/26392">PR #26392</a> (Under Review)</li><li><strong>PR 1.11 (Acceptance Tests)</strong>: <a href="https://github.com/oppia/oppia/pull/26458">PR #26458</a> (Under Review)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c8ff8a706651" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Week 6: Pushing Milestone 1 to the Finish Line and Tackling Integration Blocks]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-week-6-pushing-milestone-1-to-the-finish-line-and-tackling-integration-blocks-14dfc17a105f?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/14dfc17a105f</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[oppia]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Fri, 19 Jun 2026 10:35:45 GMT</pubDate>
            <atom:updated>2026-06-19T10:35:45.667Z</atom:updated>
            <content:encoded><![CDATA[<p>This week has been pretty busy as we get close to finishing Milestone 1. Most of my time went into integration work, connecting our backend V2 handlers with the frontend under the new generic translation system.</p><h3>Goals for this Week</h3><p>My main goals were to get PR 1.5 (Suggestion Resolver) and PR 1.7 (Pinning V2) approved, address the feedback on the large PR 1.6 (Frontend V2 Integration), and make progress on the backfill jobs.</p><h3>What I Accomplished</h3><ul><li><strong>PR 1.7 Merged and PR 1.5 Approved</strong>: PR 1.7 (Lessons Pinning Handler V2) was merged, and PR 1.5 (Suggestion Resolver V2) got approved and is now in the merge queue.</li><li><strong>Fixed the PR 1.6 Blocker</strong>: While testing the frontend with the V2 flag, we ran into an issue where the translation modal needs the exploration state name to group the cards page-by-page. To fix this without locking the V2 API to explorations, I added a generic grouping_key field. The backend passes the state name here, and the frontend groups the cards accordingly. I formatted the python code to fit the 80-character line limit and wrote the tests for it.</li><li><strong>Opened PR 1.10</strong>: Created the Beam job to backfill the entity_type field for legacy pinned opportunity records.</li><li><strong>Community Contribution GFIs</strong>: I also worked on writing breakdowns for Good First Issues (GFIs) on GitHub. I completed 4 technical breakdowns (deprecating auto-TTS, adding progress bars to the review tab, Hinglish translation tips, and SVG text parsing fixes).</li></ul><h3>What I Learned</h3><p>Decoupling exploration-specific logic from a generic backend is tricky. The state name issue showed that while we want a generic API, we still need to support exploration card grouping on the frontend. The grouping_key approach worked well because it keeps the API clean and extensible for other entities in the future.</p><h3>What I Struggled With</h3><p>The main struggle this week was managing all the open PRs at once. Between addressing code reviews on 1.5 and 1.6, writing GFI breakdowns, and preparing for the upcoming PM demo, it was easy to get overwhelmed. Branch management was also tricky since PR 1.11 (acceptance tests) is blocked until PR 1.5 and 1.6 are merged.</p><h3>Next Steps</h3><p>Once PR 1.5 and 1.6 are merged, I will start writing the acceptance tests on PR 1.11 to wrap up Milestone 1, and coordinate with the team to get the Beam jobs reviewed.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=14dfc17a105f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GSoC 2026 Week 5: Moving Fast on Controller Layers and Navigating Design Trade-offs]]></title>
            <link>https://medium.com/@rohan_unbeg/gsoc-2026-week-5-moving-fast-on-controller-layers-and-navigating-design-trade-offs-f0d169bc15e1?source=rss-ef46c971e707------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0d169bc15e1</guid>
            <category><![CDATA[gsoc]]></category>
            <category><![CDATA[oppia]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[open-source]]></category>
            <dc:creator><![CDATA[Rohan Unbeg]]></dc:creator>
            <pubDate>Tue, 09 Jun 2026 18:12:11 GMT</pubDate>
            <atom:updated>2026-06-09T18:12:11.080Z</atom:updated>
            <content:encoded><![CDATA[<p>This week was all about moving from the foundational storage and service layers into controllers and frontend services, while navigating complex design trade-offs with my mentors. With my university exams behind me, I was able to maintain a high development pace and open three different PRs!</p><h3>My Goals for this Week</h3><p>My primary goals were to get PR 1.4 (Controller layer) merged, address the remaining feedback on the PR 1.9 Beam job, open PR 1.5 (Suggestion Handler), and open PR 1.6 (Frontend Services).</p><h3>What I Accomplished</h3><ul><li><strong>PR 1.4 Merged:</strong> PR 1.4, which adds the new controller endpoints (like ContributionOpportunitiesHandlerV2 and ReviewableOpportunitiesHandlerV2) gated behind the new feature flag, was successfully approved and merged!</li><li><strong>Opened PR 1.5 and PR 1.6:</strong> I opened PR 1.5 to update the SuggestionHandler logic to resolve suggestion target IDs using the new generic translation opportunities. I also opened PR 1.6 to integrate these backend changes with the frontend services (ContributionOpportunitiesService and the API services).</li><li><strong>Refactoring PR 1.9 Content Count (Sync with Hasitha):</strong> During our GSoC sync, Hasitha and I discussed two implementation options for the get_content_count() logic in the Beam job. We chose Option 1 (updating counts to include metadata fields, gated under the feature flag). I refactored the Beam job, wrote integration tests covering both feature flag states, added automated verification to the audit job, and updated the Release Testing Doc.</li><li><strong>Ready for Backup Server Testing:</strong> With all comments on PR 1.9 addressed and local verification complete, the Beam job is now awaiting a final pass from Hasitha to begin backup server testing.</li></ul><h3>What I Learned</h3><p>This week taught me a lot about designing backward-compatible APIs during major database migrations.</p><p>The frontend suggestion review modal expects opportunity details (like topic_name, story_title, and chapter_title) to render its subheadings. Since the new TranslationOpportunity model is generic and doesn&#39;t store story or chapter fields directly, we had to implement a transitional backend bridge: query the mappings on-the-fly and map them into a temporary ExplorationOpportunitySummary dictionary.</p><p>Learning how to roll out new models incrementally while keeping the legacy frontend fully functional is a crucial real-world software engineering skill.</p><h3>What I Struggled With</h3><p>My main challenge this week was achieving 100% line and branch test coverage for the integration tests.</p><p>Testing the new controller endpoints under both enabled and disabled feature flag states required writing comprehensive integration tests in core/controllers/suggestion_test.py. Setting up the mocks (swapping get_all_topic_summaries, get_node_with_corresponding_exp_id, etc.) to cover edge cases like empty topic summaries or invalid/missing targets took several iterations to get exactly right. However, getting the green checkmark on the CI checks made the effort completely worth it.</p><h3>Next Steps</h3><p>For next week, my main focus will be executing the backup server testing for the PR 1.9 Beam job so we can get it merged. I’ll also address the review comments on PR 1.5 and 1.6 to wrap up Milestone 1!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f0d169bc15e1" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>