Akilan Manikandan

Designed the backend-oriented security flow and implemented service boundaries for encrypted file operations, sharing policy checks, and audit-friendly access.

Files move through a controlled pipeline covering upload, encryption, key exchange, policy validation, access control, audit logging, and monitored retrieval. Separate services manage storage security, sharing workflows, and anomaly detection.

It addresses insecure document sharing in distributed environments by enforcing strong access boundaries, protecting sensitive data, and improving traceability for every file-level operation.

The design separates file operations, access validation, and anomaly monitoring so the authorization model remains understandable as policies and usage grow.

Built the backend API layer and recommendation flow connecting event ingestion, normalized storage, contextual query handling, and frontend consumption.

Event data is collected through ingestion pipelines, normalized, stored in MongoDB, and exposed through FastAPI endpoints. A recommendation layer then maps user intent to relevant events using contextual query handling.

It reduces the friction of event discovery by replacing fragmented listings with a single recommendation flow that can answer user queries and return more relevant options quickly.

The system separates ingestion, storage, API delivery, and recommendation logic to avoid coupling chatbot behavior directly to raw event data.

Developed the audio preprocessing and model-training workflow, including MFCC feature extraction, LSTM experimentation, and multi-class evaluation.

The pipeline preprocesses audio, extracts MFCC features, and feeds the sequential representation into an LSTM model trained to classify emotional states from speech samples.

It enables systems to interpret emotional cues in speech, which is useful for voice interfaces, assistive technologies, and conversational systems that require more context than transcription alone.

Preprocessing, feature extraction, training, and inference are kept modular so experiments can be repeated, evaluated, and improved independently.