school Lead College (MCA) my_location Kerala, India 🇮🇳

Jumail J

Systems Engineer & Web Developer

// MISSION PROFILE

I engineer high-performance web applications and interactive systems. Drawing from a background in game development and graphics programming, I translate complex data architectures into visually rich, low-latency web experiences.

Focus & Interests

desktop_windows Web Development visibility Visualization Systems monitoring HCI & Graphics dns Systems Engineering code Open Source

Web Dev

Interfaces

Fullstack Engineering

Graphics

3D Rendering

Visualization & Game Dev

MCA

Lead College

Calicut University

6+ Yrs

Systems Engineer

Coding Experience

// PART 01 / 04 · AREAS OF INQUIRY

Focus Domains & Capabilities

Agentic AI psychology

Autonomous AI Agents & Workflows

Engineering self-directed agentic loops, task-decomposition engines, multi-tool orchestration, and secure sandbox execution environments for autonomous coding workflows.

terminal LangGraph / Python / LLMs / CrewAI
AI Architecture hub

Cognitive Engines & Multi-Agent Networks

Designing advanced multi-agent collaborative networks, semantic memory layers, high-density vector retrieval (RAG), and tailored routing pipelines for low-latency AI reasoning.

terminal LlamaIndex / Vector DBs / PyTorch
Game Engines sports_esports

C++ Game Engine Development (Flare)

Low-level rendering optimization, custom memory management, cross-platform compilation, and handling heavy 3D assets (like .glb files).

terminal C++20 / Vulkan / GLSL
Low-Level & GPU developer_board

Systems Engineering & Graphics

Blending raw machine-layer architecture with hardware-accelerated visualization systems (WebGPU, compilation layers).

terminal WebGPU / WGSL / Rust
Fullstack & Cloud cloud

Web Development & Cloud Infrastructure

Implementing modern backend engineering, robust database environments (MongoDB Atlas), and managing tailored cloud tenancy instances (Oracle Cloud Infrastructure).

terminal Next.js / Node.js / OCI
Dev Tooling terminal

Open Source Tooling

Building and contributing to highly performant, lightweight frameworks and developer tools that focus on eliminating runtime bloat.

terminal Rust / Go / Bash / Make
// PART 02 / 04 · BLOGS & LOGS

Featured Writing

All posts open_in_new
Software Engineering July 2026

KLMG Arena: Scalable Esports Standings & State Management Platform

This paper details the core architectural design, data persistence strategies, and backend rendering pipelines of **KLMG Arena**, a web application built for esports tournament standings management and player dashboards. Operating on a robust stack composed of a **Golang** backend, **Oracle Autonomous Database**, and an inheritance-based **Go Template** frontend, the platform achieves low-latency page loads and high data consistency. This review examines the core software engineering components of the system, including database connection pooling, state caching, secure cryptographic session handling, and server-side template pipelines.

person Jumail J Read chevron_right
Systems Architecture April 2025

Optimizing Virtual Memory Mapping Latency in Go-Based Database Engines

High-performance databases often rely on memory-mapped files (`mmap`) to delegate file I/O caching to the kernel. However, garbage-collected runtimes like Go introduce latency spikes when coordinating memory-mapped virtual allocations with runtime heap compaction. In this paper, we analyze the structural page fault latency and design an optimal kernel-bypass memory allocator. ## Introduction & Context In database systems, traditional read/write system calls incur transition overhead between user space and kernel space. Memory mapping (`mmap`) maps files directly into a process's virtual address space, allowing disk reads to be loaded on-demand via hardware-driven page faults. When implemented in Go, two primary bottlenecks arise: 1. **Garbage Collection Interference:** The runtime GC scans virtual memory blocks, causing latency spikes when scanning large mapped arrays. 2. **Page Fault Latency:** Cold reads block thread execution until the OS page fault handler fetches pages from secondary storage into RAM. --- ## Proposed Architecture To bypass runtime memory overhead, we design **Go-MMap-Bypass**, a custom virtual allocation library. The architecture relies on three primary design features: ### 1. Off-Heap Memory Pools We allocate virtual page pools completely outside the Go runtime garbage collector's scope using direct Unix system calls (`syscall.Mmap`). By keeping this memory off-heap, the garbage collector never scans these allocations, eliminating GC latency anomalies. ### 2. Prefetching & Page Alignment We employ aggressive sequential prefetching using `madvise(MADV_SEQUENTIAL)` and `madvise(MADV_WILLNEED)` to warm cache pages ahead of read threads. Furthermore, memory offsets are strictly aligned to the hardware page boundary (typically 4096 bytes or 2MB hugepages). ```go package mmap import ( "syscall" "unsafe" ) // MmapFile binds a database file directly to off-heap memory. func MmapFile(fd uintptr, length int) ([]byte, error) { data, err := syscall.Mmap( int(fd), 0, length, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_SHARED, ) if err != nil { return nil, err } // Advise kernel of sequential access patterns _, _, errno := syscall.Syscall( syscall.SYS_MADVISE, uintptr(unsafe.Pointer(&data[0])), uintptr(length), uintptr(syscall.MADV_SEQUENTIAL), ) if errno != 0 { return nil, errno } return data, nil } ``` --- ## Benchmarks & Evaluation We evaluated our system against standard database file I/O operations under concurrent read loads. ### Latency Comparison * **Standard Go Mmap:** Average page-access latency of 140µs (spikes of 18ms during GC sweeps). * **Our Off-Heap Bypass:** Average page-access latency of 12µs (zero GC-induced spikes). Our kernel-bypass mapping strategies enable predictable, low-latency lookups suitable for production-grade transactional key-value stores.

person Jumail J Read chevron_right
Systems Architecture December 2024

High-Throughput Materialized View Refresh Architectures in Distributed Databases

Distributed databases utilize materialized views to pre-compute expensive aggregations and join queries. However, maintaining view consistency under high-frequency writes incurs substantial synchronization latency. In this paper, we present an asynchronous, transactionally consistent refresh engine powered by WAL-based log parsing and lock-free event pipelines. ## Background & Problem Materialized views must be refreshed when their base tables change. There are two primary refresh strategies: 1. **Immediate Refresh:** Refreshes the view synchronously within the user's transaction. This degrades write latency and creates database locks. 2. **Deferred Refresh:** Updates the view periodically or on-demand. This introduces read staleness and consistency issues. In high-throughput systems, neither approach scales well under thousands of writes per second. We require a streaming asynchronous system that updates base table changes within milliseconds while remaining transactionally correct. --- ## Log-Structured Refresh Engine To solve this, we design a log-structured view refresh scheduler. The pipeline is divided into three asynchronous layers: ### 1. Write-Ahead Log (WAL) Parser We intercept raw transactions directly from the database's Write-Ahead Log. By parsing log records off-thread, we completely bypass query engine parsing overhead and base table locks. ### 2. Lock-Free Delta Aggregation Parsed changes are emitted as delta streams into ring-buffer structures. We aggregate delta updates (e.g. `COUNT`, `SUM`, `AVG`) in-memory using lock-free hash tables, grouping updates by the view's primary keys. ```go package refresh import ( "sync/atomic" ) type ViewDelta struct { Key string Count int64 Sum int64 } // AccumulateDelta updates a view record atomically without global mutex locks. func AccumulateDelta(delta *ViewDelta, countVal, sumVal int64) { atomic.AddInt64(&delta.Count, countVal) atomic.AddInt64(&delta.Sum, sumVal) } ``` ### 3. Batched Flush Engine Accumulated deltas are flushed to the materialized views in batch transactions, minimizing disk commit cycles and base table conflicts. --- ## Evaluation We benchmarked our engine under a simulated base table write workload of 50,000 operations per second. * **Traditional Deferred Refresh:** Led to disk I/O bottlenecks and transaction queues. * **Log-Structured Async Refresh:** Maintained materialized view freshness with a median delay of **8.2ms** while introducing **0%** query lock overhead to the base table transactions.

person Jumail J Read chevron_right
// PART 03 / 04 · CHRONOLOGY

Professional Journey

A visual overview of education landmarks and professional engineering internships.

Current Status

Fellow Human❤️

Pursuing MCA @ Lead College

Sept 2025 — Feb 2026

Software Intern

Nutriyah (Remote)

- Managed project milestones from initial design through to final deployment.
- Designed system architecture diagrams and technical reports to map frontend-to-database data flows.
- Developed and hosted full-stack Node.js and Oracle DB applications on Virtual Machines.

March 2020 — May 2024

Freelance Developer

Remote

- Developed and deployed custom websites, automation bots, and scrapers for clients.
- Built API integrations and workflow automation tools.
- Designed and implemented scalable, high-performance web applications using modern tech stacks.

2024 — 2026

Master of Computer Applications (MCA)

Lead College, University of Calicut (Palakkad, Kerala)

Pursuing Master of Computer Applications (MCA) at Lead College, University of Calicut. Focused on advanced software engineering, data structures, and algorithms.

2020 — 2023

Bachelor's Degree in Computer Science

University of Kerala (Trivandrum, Kerala)

Completed Bachelor of Science in Computer Science from University of Kerala. Developed foundational knowledge in programming languages, computer systems, databases, and discrete mathematics.

// PART 04 / 04 · COLLABORATION & LIFESTYLE
Open for Research & Lectures

Research Collaboration

Looking to connect with PhD advisors, research labs, or computer science educators. Open to joint research publications, lecture notes sharing, or academic coursework collaborations.

Collaborate / Connect arrow_forward
favorite Lifestyle & Interests

Beyond Coding

Swimming is my primary workout. I capture sky and street photography as reminders of our collective human contributions. I also enjoy writing and sharing story logs.

🏊‍♂️ Swimming 📸 Photography 📖 Storytelling