AVAILABLE FOR REMOTE ROLES (IMMEDIATE JOINER · UTC/EST/CET OVERLAP)

Shahriar Mehedi

Full-Stack & AI Systems Engineer

I architect and deploy fault-tolerant web applications, autonomous AI workflows, and edge-to-cloud telemetry systems. Focused on predictable performance, low-latency data pipelines, and lean infrastructure unit economics.

↗ Resume (PDF)↗ Book 15-min Intro ↗ GitHub ↗ LinkedIn

▶ Watch 90s Architecture & System Design Intro (Loom) ↗

02 // SYSTEMS

01 / Pro Tasker — Full-Stack Productivity Platform & AI Workspace

Live System (1-Click Guest) ↗Source ↗

Next.js 15 · TypeScript · MongoDB Atlas · Prisma 6 · Gemini 2.5 Flash · NextAuth · PWA · SSE

Production-grade SaaS productivity platform combining natural-language task planning, Pomodoro focus tracking with Picture-in-Picture, collaborative canvas drawing, and real-time presence.

• Bottleneck:Direct client state and simultaneous AI task generation caused UI race conditions, duplicate writes, and WebSocket connection overhead.
• Architecture:Engineered a layered React Context architecture with optimistic state mutations, temporary ID reconciliation on server confirmation, atomic rollback on failure, and Server-Sent Events (SSE) for real-time presence over standard HTTP/2 streams.
• Impact & Unit Economics:Delivered 35+ features across an installable PWA with sub-80ms UI responsiveness, offline service worker caching, and zero dedicated WebSocket infrastructure costs ($0/mo real-time layer).

Topology:

[Edge Client Stream] ──► [Ingestion Gateway] ──► [Redis Streams / BullMQ]
                                                        ├──► [Gemini 2.5 Planner] ──► [Structured Tasks]
                                                        └──► [SSE Presence Hub]  ──► [Live Collaborators]

02 / Semantic ATS — AI-Powered Applicant Tracking System

Live System (1-Click Guest) ↗Source ↗

Next.js 16 · PostgreSQL · pgvector · BullMQ · Redis · Gemini AI · Prisma ORM · TypeScript

Full-stack recruitment intelligence platform that replaces broken keyword matching with high-dimensional 3072-vector semantic similarity search powered by Google Gemini and pgvector.

• Bottleneck:Serverless Vercel function timeouts during PDF tokenization and cascading Gemini API 429 rate limits during bulk candidate ingestion.
• Architecture:Decoupled Next.js ingestion from a standalone BullMQ worker daemon on Railway backed by Redis; throttled jobs to 10/min; stored 3072-dim embeddings in PostgreSQL using the pgvector extension via raw parameterized SQL (<=> cosine distance).
• Impact & Unit Economics:Scored zero-keyword-overlap candidates with 75–90% conceptual accuracy, cut HR screening time by 60%, and eliminated external vector database SaaS costs ($0 extra hosting cost by keeping vectors in PostgreSQL).

Topology:

[Client / PDF Upload] ──► [Next.js API Gateway] ──► [BullMQ Task Daemon]
                                                          ├──► [Gemini text-embedding-004] ──► [3072-dim Vector]
                                                          └──► [PostgreSQL + pgvector]     ──► [Cosine Distance <=>]

03 / IntelliTex — Premium Minimalist LaTeX Editor & AI Copilot

Live System (1-Click Guest) ↗Source ↗

Next.js 16 · React 19 · Monaco Editor · MongoDB Atlas · Prisma ORM · Gemini AI · Tailwind CSS v4

Cloud-synced, Monaco-powered LaTeX editor and real-time previewer with debounced local-first sync to MongoDB Atlas, a secured Gemini AI copilot with live editor context, and rolling restore snapshots.

• Bottleneck:Imperative Monaco Editor listeners capture stale React closures on mount; re-registering handlers on render leaks disposables and creates duplicate hotkey bindings.
• Architecture:Stored active document content in a mutable useRef cache accessed by a mount-time command handler; debounced cloud writes to MongoDB Atlas (1500ms) with Ctrl+S immediate bypass; secured Gemini AI copilot via a server-side proxy route.
• Impact & Unit Economics:Achieved 0ms input typing lag, 15 rolling compilation restore points, and zero hydration flicker, running entirely within serverless free-tier limits ($0/mo infrastructure spend).

Topology:

[Monaco Keypress (0ms)] ──► [Mutable useRef Cache] ──┬──► (Debounced Save 1.5s) ──► [MongoDB Atlas]
                                                    └──► (Ctrl+S / AI Copilot) ──► [/api/copilot Proxy] ──► [Gemini 2.5]

03 // EXPERIENCE

2021 — PRESENTRemote

Full-Stack Software Engineer (Independent / Contract)

•Architected and deployed 15+ production web applications using Next.js, Node.js, PostgreSQL, and MongoDB with automated Docker & CI/CD pipelines.
•Standardized secure authentication flows utilizing HttpOnly cookies, rotating refresh tokens, and role-based access control (RBAC).
•Built autonomous AI agent workflows, data ingestion pipelines, and custom automation tooling using TypeScript and Python.

2023 — PRESENTRajshahi, BD

B.Sc. in Computer Science & Engineering

•Researching edge-computing and AI architectures for automated hydroponic nutrient management (IEEE publication format) and low-level system design.

04 // STACK

LANGUAGESTypeScript, JavaScript (Node.js), Python, SQL (PostgreSQL), C/C++ (ESP32), x86 Assembly

FRONTENDNext.js (App Router), React 19, Tailwind CSS, Server-Sent Events (SSE), WebSockets, Monaco Editor

BACKEND & DATAPostgreSQL, MongoDB Atlas, Prisma ORM, Redis (Streams & Cache), BullMQ, RESTful APIs

INFRA & AIDocker, Linux VPS, Nginx, GitHub Actions (CI/CD), LLM Orchestration, pgvector (RAG)

05 // CONTACT

Available for immediate full-time remote engineering roles and architectural consulting.

Schedulecal.com/shahriarmehedi (15-min intro call) ↗

md.shahriar.mehedi@gmail.com

Profiles

GitHub ↗·LinkedIn ↗·Resume (PDF) ↗