PREM MALLAH · AI/ML ENGINEER

Computer vision.
Built for production.

I build video inference pipelines and connect AI detections with event processing, security hardware, and operator-facing services. Currently a Jr. AI Developer at Closyss Technologies, based in Mumbai, India.

PRODUCTION AI COMPUTER VISION DISTRIBUTED SYSTEMS AWS KAFKA GPU INFERENCE HARDWARE SIA DOCKER
01 / SELECTED WORK

SMART ALERT OPERATIONS

Turning high-volume security and video streams into a coordinated operational workflow — combining real-time vision inference, event routing, and AI-assisted operator review.

MY CONTRIBUTION
Built the end-to-end alert workflow, computer-vision inference services, event-processing pipeline, and cloud-to-edge infrastructure.
SYSTEM OUTCOME
Reduced alert noise and turnaround time through automated event correlation, balanced operator routing, and AI-generated review context.
01

The Operational
Problem

High-volume monitoring environments generate continuous streams of raw signals and isolated triggers. Processing every event independently creates operator fatigue and delays response times. The goal was to replace fragmented alert handling with an intelligent, coordinated pipeline.

02

Intelligent
Alert Workflow

  • Event correlation. Consolidate related triggers within active time windows to eliminate duplicate noise.
  • Workload-aware routing. Distribute incoming incidents dynamically across available operators.
  • Automated continuity. Handle active queues seamlessly across operational transitions.
  • AI-assisted context. Attach automated visual analysis and concise summaries before human review.
03

High-Level
Pipeline

Live video and hardware signals are ingested from remote sites, analyzed by vision models, and routed through an event-driven backend using Apache Kafka, WebSockets, and REST APIs to deliver low-latency updates to operators.

04

Computer Vision
& Edge Integration

Trained and deployed PyTorch and YOLO vision models for continuous real-time detection over live RTSP streams, while integrating physical security hardware with backend services through standard industry protocols.

05

Cloud & Edge
Infrastructure

Containerized services are deployed via automated CI/CD pipelines across cloud environments (AWS, Azure, OCI) alongside on-premise GPU servers configured for sustained video inference.

* Architecture and workflow details are generalized to respect proprietary systems.

Experience

2025 — PRESENT

CLOSYSS TECHNOLOGIES

Jr. AI Developer
• Engineered production computer-vision pipelines for real-time event detection.
• Designed distributed event-processing components using Kafka and real-time APIs.
• Integrated AI inference with distributed backend services and hardware communication layers.
Selected focus: Computer vision inference · Kafka pipelines · Dockerized services · cloud infrastructure · edge hardware integration
02 / INTERACTIVE EXPERIMENTS

PREM'S LABS

A collection of interactive computer-engineering laboratories. Explore systems architecture, networking, and bit-level concepts through visual, client-side simulations.

ENTER THE LABS →

Technical Stack

AI / MACHINE LEARNING

  • PyTorch
  • YOLO
  • ONNX
  • TensorRT

SYSTEMS & DATA

  • Docker
  • Apache Kafka
  • PostgreSQL
  • Linux

CLOUD

  • AWS
  • Azure
  • OCI

VIDEO & NETWORKING

  • RTSP
  • WebRTC
  • HLS
  • WebSockets

LANGUAGES

  • Python
  • Java
  • JavaScript
  • SQL

CERTIFICATIONS

  • AWS Solutions Architect (SAA-C03)
  • OCI Architect Associate (2025)
  • OCI Networking Professional (2025)
  • Azure Fundamentals (AZ-900)
  • NVIDIA Deep Learning

About

I'm Prem — an AI/ML engineer interested in the space where models meet real systems.

I spend most of my time working on computer vision, production inference, distributed services, edge hardware, and the infrastructure required to keep these systems running reliably.

Currently building production AI systems at Closyss Technologies.

LET'S BUILD SOMETHING USEFUL.

For professional inquiries regarding AI/ML engineering, distributed systems, or applied research: