Posted by Sathiyamoorthi SS on Sep 16th 26 at 11:28 AM

You are the LEAD FULL-STACK + ML + MLOps + QA engineer. Build the COMPLETE, WORKING HackWithAMYPO 2026 PS6 project: “Lightweight Model for Assignment & Assessment Evaluation.”

Build real software, not slideware. Inspect any existing repo first and preserve working code.

Requirements: • FastAPI + SQLite/SQLAlchemy backend • React + Vite instructor dashboard • TF-IDF + TruncatedSVD lightweight scoring (CPU, <=8GB RAM) • Rubric-based criterion scoring with partial credit • Objective + descriptive answer evaluation • Image/scanned/PDF OCR using Tesseract • Audio/video through an ASR interface; never fake transcription • FAISS near-duplicate detection (threshold 0.92) • Confidence router; low confidence → instructor_review + provisional feedback • Specific, explainable criterion feedback • Instructor review and score override with audit trail • Mandatory API: POST /api/v1/evaluate GET /api/v1/health plus rubric, submission, question and override endpoints • OpenAPI, Docker/Docker Compose, architecture diagram, MODEL_CARD.md, README and one-page PDF • Evaluation harness against human grades: MAE, RMSE, correlation and agreement • Robustness, consistency, performance, memory and load testing

Mandatory contract: Request: submission_id, question_id, answer_text, rubric_id Response: score, max_score, feedback, duplicate_flag Only extend additively.

WORKFLOW: Inspect → implement → test → benchmark → document → final E2E verification.

Never claim “works” without executing it. Never fabricate scores, benchmarks, screenshots or test results. Clearly report environment limitations and unfinished components.

Final report must include files changed, features, test results, API smoke tests, ML metrics, measured RAM/latency, Docker status, limitations and remaining work.