Aryan Anand

Ninestars Information Technologies, Bangalore

• Architected and developed AOTM Vault, a multi-tenant secrets management platform supporting 50+ cryptographic algorithms, multi-cloud storage across AWS, Azure and Google Cloud, automated key rotation, real-time WebSocket streaming and immutable audit logging, reducing enterprise key rotation overhead by 80%.
• Architected AOTM FDI, a multi-tenant credential-fraud detection platform with 11 Django apps, 119 REST routes and a 43-table PostgreSQL schema; built a 5-stage OCR-to-clustering pipeline using RabbitMQ and specialised Celery workers with idempotent distributed processing, WebSocket progress streams, collaborative editing and report generation.
• Engineered high-scale data and platform systems including a distributed migration engine for 60M+ row databases, a Redis/PostgreSQL usage and billing layer with a sub-5ms hot path, and fail-closed multi-tenancy across 33 tenant-scoped tables; shipped 170+ versioned builds in seven weeks backed by 70 scripted test suites.
• Built distributed ML infrastructure across 10 RabbitMQ quorum queues with transactional fan-in and redelivery-safe processing, executing 10,700+ inference jobs across 2,500+ completed multi-stage runs; designed telemetry systems sustaining 3.6M rows over 119 days and serving 50K-row analytical windows in ~250 ms.
• Built an enterprise RAG system using hybrid dense+sparse search over Qdrant with self-hosted LLMs and Jina Embeddings, enabling sub-800ms semantic search across 100K+ files; also designed a private blockchain ledger with immutable transaction chains, Merkle-root validation and on-chain document signing.

TIH IoT, IIT Bombay, Mumbai

• Built a U2-Net image masking pipeline and CNN-based vegetation segmentation system for NDVI and hyperspectral imagery, achieving an 11% improvement through preprocessing, augmentation and hyperparameter tuning.

Python, PyTorch, TensorFlow, Llama, React.js, Node.js, MongoDB, Git

• Built a super-resolution GAN for MRI/CT clarity achieving 92% accuracy on 54,000 samples, a CycleGAN for MRI-to-CT translation, and a Vision Transformer-CNN for multi-label disease classification achieving 88.4% accuracy; fine-tuned Llama 3.2 11B on medical literature for explainable AI insights.

Python, OpenCV, TensorFlow, Google Cloud, Git

• Developed a real-time sign language-to-speech system using ResNet-18 and OpenCV achieving 97% accuracy and 200ms latency on low-end devices, with Google TTS API integration.