The Mind of Sommayadeep

Feature Importance

turbofan-rul-prediction

Problem: Aviation maintenance needed earlier failure signals from high-dimensional sensor streams.

Challenge: Sensor drift and noisy operating conditions made RUL models unstable.

Solved: Built an ensemble pipeline with feature selection, robust scaling, and model blending.

Tradeoff: Accepted slower training runs to gain better generalization on unseen engine units.

CertiTrust

Problem: Institutions needed tamper-proof certificate verification without manual checking.

Challenge: On-chain transparency had to coexist with privacy-safe document handling.

Solved: Stored deterministic hashes on-chain with lightweight role-gated verification APIs.

Tradeoff: Kept metadata minimal on-chain to reduce gas while preserving verification integrity.

student-management-system

Problem: Schools needed one role-safe system for attendance, grades, and eligibility decisions.

Challenge: Concurrent updates from teachers/admins caused inconsistent attendance records.

Solved: Designed role-scoped services with atomic attendance updates and audit-safe grade flows.

Tradeoff: Introduced stricter backend validation rules, adding small write latency for consistency.

The Challenge

Sensor streams had noisy dimensions and engine-condition drift, causing unstable Remaining Useful Life predictions.

The Solution

Built a robust pipeline with feature pruning, scaling, and blended models (Random Forest + HistGradientBoosting) to handle non-linear behavior.

The Result

Improved validation MAE by 18% over baseline, reduced noisy dimensions by 35%, and kept inference under 120ms for batch scoring.

Deployment Architecture

CI/CD Flow

1. 1. Commit + Pull Request
2. 2. Type Check + Lint + Build Validation
3. 3. Preview Deployment + Smoke Checks
4. 4. Main Branch Merge
5. 5. Production Deploy + Health Monitoring

Simulated Benchmarks