Adaptive Reinforcement-Based Personalized Worksheet Generation - #92
Open
vanshikaagrawalcsaiml24 wants to merge 142 commits into
Open
Adaptive Reinforcement-Based Personalized Worksheet Generation#92vanshikaagrawalcsaiml24 wants to merge 142 commits into
vanshikaagrawalcsaiml24 wants to merge 142 commits into
Conversation
…safe
The diagnostic-submit handler shelled out to the Python pipeline in a way
that could never succeed and was unsafe:
1. Wrong interpreter — it invoked `python`, which doesn't exist on the
deployment (Ubuntu 24.04 ships python3, no bare `python`), so every run
threw "python: not found" and silently fell back to Gemini.
2. Wrong directory — cwd was `<backend>/evaluation_metrics`, which does not
exist; the pipeline lives in the sibling `ai-services/` (renamed in the
restructure, never updated in code). So even a correct interpreter would
fail on a missing cwd.
3. Command injection — `execSync` ran a shell string with `classNumber` and
`student.id` interpolated in, so any crafted value would be shell-executed.
Fix:
- Introduce PYTHON_BIN (default `python3`) and AI_SERVICES_DIR (default the
sibling ai-services/), both env-overridable.
- Point the pipeline cwd at AI_SERVICES_DIR.
- Replace both execSync shell strings with execFileSync + array args (no shell),
so classNumber/student.id are passed literally and cannot inject.
Verified on the target host: `execFileSync('python3', ['run_pipeline.py', ...],
{cwd: ai-services})` now launches python3 inside ai-services and reaches the
script's own logic (no ENOENT for interpreter or cwd).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ipeline-exec fix: make Python evaluation pipeline runnable (python3 + correct dir) and injection-safe
PDF generation (diagnostic papers, worksheet rendering) launched Puppeteer with no usable Chrome: CHROME_EXECUTABLE_PATH was unset, Puppeteer's managed Chrome was never (fully) installed, and there was no system Chrome — so puppeteer.launch() threw at runtime and every PDF request 500'd. A plain `npm install` also failed on the browser download. - Add backend/src/browser.ts: a single launchBrowser() helper used by both paperGenerator.ts and worksheetRenderer.ts (removes the duplicated launch config). It honors CHROME_EXECUTABLE_PATH and, on failure, throws one clear, actionable error (install via `npx puppeteer browsers install chrome` / system libs / set the env var) instead of a raw Puppeteer stack. - Document CHROME_EXECUTABLE_PATH and the browser-install step in .env.example. Provisioning is environment setup (Option A): install the version-matched Chrome with `npx puppeteer browsers install chrome` plus its system libraries. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…r-chrome-setup fix: reliable headless-Chrome launch for PDF generation
frontend/src/utils/levelGenerator.ts was a byte-identical copy (except its import line) of backend/src/levelGenerator.ts — ~485 lines that generate the question sets AND their correct answers for all 59 levels. It is now dead code: nothing in frontend/src imports it (a leftover from the deleted in-browser mock, which used to generate questions client-side; the frontend now gets questions from the API). Keeping it risked (a) silent drift from the backend copy, and (b) leaking every question's answer key into the client bundle if it were ever imported again. Delete it, leaving backend/src/levelGenerator.ts as the single source of truth. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…up-levelgenerator refactor: remove unused duplicate frontend levelGenerator
…ching)
Every API and asset URL in the frontend was hardcoded root-absolute
(`fetch('/api/...')`, `'/worksheets/x.html'`). At the domain root that works,
but under the /fln subpath a page at tenali.fun/fln/ calling fetch('/api/x')
resolves to tenali.fun/api/x — escaping the subpath. That's why the deploy had
to string-rewrite every built file (apply_fln_subpath.js) on each release.
- Add src/services/apiClient.ts: withBase(path) + apiFetch(path, init), keyed off
import.meta.env.BASE_URL (Vite's base: "/" at root, "/fln/" for the subpath
build). Trailing slash handled so there's no "//".
- Route all 62 fetch('/api...') calls through apiFetch, and the worksheet
iframe/href asset URLs through withBase.
- Make the axios client (services/api.ts, used by coordinatorService) base-aware
too (baseURL default withBase('/api') instead of a wrong localhost URL).
- vite.config: base = process.env.VITE_BASE_PATH || '/', so a subpath build just
sets VITE_BASE_PATH=/fln/ — no --base flag and no built-file rewriting needed.
The same source now works at the root and under any subpath with zero patching.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…re-api-urls fix: base-path-aware API/asset URLs (support /fln subpath without patching)
…only export error
…ne and prevent question reuse
…ness tiers, and added PDF reinforcement badges
…ger rules, and level+2 targeting
…ent, 4-question PDFs, and verification panel alignment
…ing original 4 level questions
…top, non-repeating variants, and teacher alerts
…rksheet composition, and multi-concept support - Score <40%: reinforce every worksheet - Score 40-69%: reinforce every alternate worksheet - Score >=80%: stop reinforcement - 0 weak concepts: 5 normal questions - 1 weak concept: 4 normal + 1 reinforcement - 2 weak concepts: 3 normal + 2 reinforcement - 3+ weak concepts: 2 normal + 3 reinforcement (top 3 weakest) - Reinforcement questions are unique per worksheet and per concept - Continue until concept reaches >=80% mastery or 3 cycles complete - All 14 tests passing
…sheet and enforce score >= 80% mastery stop rule - Score >= 80% (e.g. Number Operations 84%) now immediately deactivates reinforcement, marks concept ineligible (questionsToInject: 0), and stops reinforcement. - Score <40% reinforces every worksheet; 40-79% reinforces every alternate worksheet. - Reinforcement Verification Panel debug info is now generated atomically with getReinforcementQuestionsWithDebug, ensuring 100% synchronization between panel items and actual worksheet questions. - All 15 unit tests passing.
…lation, deterministic seeding, and 100% panel-worksheet synchronization - Use latest assessment score only for mastery percentage calculation (ignore old historical averages). - Ensure 100% synchronization between Reinforcement Verification Panel and generated worksheets/PDFs. - Enforce strict score >= 80% deactivation (mastered concepts automatically removed from active weak list). - Implement deterministic PRNG question generation for student state. - Update UI panel to render Inject badge only when questionsToInject > 0. - All 11 unit tests passing cleanly.
…, atomic single-pdf debug execution, and 12 passing unit tests)
…roduce Remedial Intervention Required status - A concept can show Needs Practice (score < 50%) or Satisfactory (score 50-79%) in the scorecard. - Upon completing 3 reinforcement cycles without reaching 80% mastery, concept status becomes Remedial Intervention Required. - Concepts with status Remedial Intervention Required or score >= 80% (Strong) are immediately deactivated and marked NOT eligible for further reinforcement. - Scorecard, Reinforcement Verification Panel, Trigger Worksheet, and Generated PDFs remain 100% synchronized. - Verified all edge cases, 12 unit tests passing, zero TypeScript errors in backend and frontend.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
11 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Research-Informed Adaptive Reinforcement Framework for FLN | AI-powered personalized worksheet generation based on concept-level performance, adaptive reinforcement, mastery tracking, teacher intervention, and evidence-based learning principles including Spaced Repetition and Retrieval Practice.