iCloudEMS
Website:
icloudems.com
Job details:
The problem you are solving
250 million students sit exams in India every year. Every answer sheet is graded by a tired human with a red pen, in a room with no AC, in 45-degree heat, under political pressure. The variance in marks between two evaluators on the same paper can be 15-20%. That is not education. That is luck. We are building the AI that changes this — reading handwriting, understanding context, scoring semantically, flagging uncertainty, matching faculty-level accuracy. We need the engineer who will make this work at scale.
What you will build
HTR
Handwriting recognition on smudged, mixed-script answer sheets — not clean printed text
Layout
Detect question boundaries, diagrams, tables, margin notes across 20+ page booklets
NLP
Semantic scoring — does this answer mean the right thing, not just contain the right words
Confidence
Know when the model is unsure and route to human review with explainability
Scale
Process millions of pages during exam seasons — async pipeline on AWS SQS + GPU workers
Must have built something hard — any of these
Document OCR / HTR pipeline
Fine-tuned LLM for domain scoring
Layout / document understanding model
Semantic similarity / NLI at scale
GPU inference pipeline (TensorRT/ONNX)
Multi-language NLP (Indic languages +)
Competition — 3 rounds
Round 1
Given 50 real (anonymised) answer sheets: build a pipeline that extracts, structures and scores them. Accuracy vs faculty marks is the metric.
Round 2
48-hour sprint: improve your Round 1 model given live feedback on failure cases. Confidence calibration included.
Round 3
Live oral defence with Sachin and Hemant. Walk us through every architectural decision. We will break it.
Click on Apply to know more.