Gradence is more than just feedback. We provide the complete infrastructure for assessment: precise scoring, structured analysis, and shareable reports that prove long-term growth.
We standardize the heavy lifting. You stay in control.
Supporting 130+ students across 13 active classes (elementary to high school), with planned scale-up to 800+ students.
We don't just grade essays. We digitize the entire assessment workflow.
We bridge the analog-digital divide. Our computer vision pipeline extracts handwritten text from paper exams with high fidelity, distinguishing between student writing and printed prompts.
From IELTS speaking tests to classroom presentations. We perform automated transcription followed by rhetorical analysis to grade fluency, pronunciation, and coherence.
We support the files schools actually use. Native processing for HWP (Hangul Word Processor), Microsoft Word, and PDFs ensures no student is left behind by compatibility issues.
Gradence is architected for longitudinal growth. We track student performance from Elementary (A1) to University Entrance (C2), providing schools with a 'Golden Dataset' of student progress that spans a decade of academic development.
Analysis: Class 3B has shown a 14% improvement in Future Tense accuracy.
Querying 45 submissions from Jan 1st - Feb 1st...
Key strengths identified in speaking tasks, though written application remains consistent with previous baselines.
Forget complex pivot tables and stale spreadsheets. Our RAG Engine indexes every student submission, score, and feedback comment, allowing you to query your entire institution's performance using natural language.
We leverage a distributed, event-driven architecture designed for continuous use in school environments, handling bursts of concurrent submissions during exam periods alongside sustained background analysis for progress tracking.
Cloud credits will support our active pilot deployments and scale testing across multiple classrooms, ensuring we maintain real-time feedback performance as user load increases.
async def process_submission(file_stream): # 1. Async Ingestion (Cloud Storage) blob = await storage.upload(file_stream) # 2. Celery Task Distribution job = analyze_task.delay(blob.id) # 3. Multi-Stage Inference if blob.is_video: transcript = await speech_engine.transcribe(blob) analysis = await llm.analyze(transcript, context=history) # 4. Real-time Push socket.emit('update', {'status': 'complete', 'data': analysis})
Architectural justification for high-performance compute and enterprise-grade cloud resources.
Our pipeline performs 14 distinct LLM calls per document to ensure granular analytical depth, far exceeding standard "one-shot" grading systems.
We utilize NVIDIA-accelerated compute for Handwritten OCR (HWP & PDF) and Multimodal Video Processing to keep feedback latency under 10 seconds for our 800-student pilot.
Our "Holly" AI agents utilize cross-encoders for pedagogical precision, requiring dedicated inference workers for real-time vector retrieval.
Migrate to Gemini 1.5 Pro's 2M context window for deep-document analysis. Establish baseline cost-per-submission monitoring.
Deploy NVIDIA-accelerated Whisper-v3 for real-time oral assessments. Launch secondary pilot in Vietnam.
Evaluate transition to self-hosted Llama-3 (70B) on NVIDIA-accelerated instances for proprietary pedagogical fine-tuning.
All data is encrypted with AES-256 at rest and TLS 1.3 in transit, ensuring banking-grade security for student records.
Automated scrubbing of student data (PII) from artifacts before model inference to ensure anonymity.
Architected for FERPA, GDPR, and Korean PIPA compliance.
Support for region-locked data residency in South Korea and Vietnam.
Our models are calibrated against the IELTS 9-band scale and CEFR proficiency levels, providing validated pedagogical feedback for international school standards.
MYP Criteria A-D & DP
9-1 Grading Scale
A1 - C2 Proficiency
Standardized Exam Prep