Haris Zahid

Washmen, the largest dry-cleaning and laundry platform in the UAE, needed to scale operations while maintaining service quality, managing complex logistics, and keeping payment and support workflows smooth across a growing base of 50,000+ repeat monthly users. The core engineering challenge was that a monolithic architecture couldn't scale each part of the business independently, since billing, orders, catalog, and promotions all had different load patterns and failure modes. I redesigned the platform into a microservices architecture with each of those as dedicated, independently deployable services, which meant the team could scale and ship changes to one area without risking the others. I integrated Checkout.com for payments, handling tokenization and payment links with idempotent transaction handling so retried requests couldn't double-charge a customer, and built out AWS Lambda/API Gateway/SQS/SNS integrations for the cloud-native processing layer underneath it. To keep the system reliable under high-traffic workloads, I introduced asynchronous processing patterns and strengthened observability with structured logging, monitoring, and Sentry, cutting the time it took to detect and resolve production incidents.

Long-form documents are often impractical to read when they're needed, and users have no reliable way to turn that content into something they can consume passively. I designed the back-end architecture (Django, Node.js, PostgreSQL) around the core engineering constraint of this system: documents vary wildly in length and complexity, so the pipeline couldn't assume uniform processing time. I built a queue-based, asynchronous architecture on AWS SQS/SNS that decouples ingestion from processing, taking a document through OCR-based text extraction, LLM-driven summarization, and voice synthesis — with built-in fault tolerance and self-healing, so a failure at any stage didn't take down the whole job. I built automation workflows using Selenium for task orchestration, managed the media pipeline with FFmpeg, and supported front-end integration with Next.js and React Query, including webhook-based delivery so users are notified the moment their audio is ready rather than polling for status.

Companies needed reliable access to reviews and social mentions scattered across 150+ platforms, but building and maintaining that data infrastructure in-house is complex and breaks constantly as source platforms change their formats and APIs. I built the complete Social Media Scraper UI, including Search by URL and Callback URL functionality for webhook-based JSON delivery direct to a customer's own endpoint. I built the Review Scraper module UI and integrated its back-end APIs, then developed Job History and Usage dashboards with advanced filtering and analytics for API usage and credit consumption, since customers needed visibility into what they were being charged for. I introduced client-side caching with SWR to cut redundant API calls and improve responsiveness, and built an admin-side User Impersonation feature, so support staff could troubleshoot customer accounts securely without ever needing their credentials. I also implemented compatibility layers supporting both V1 and V2 API responses during a platform migration, so existing customers didn't experience breaking changes mid-transition.

Fashion retailers needed a scalable B2B platform to source products, manage inventory, and transact with vendors and manufacturers across a large, growing catalog. The engineering challenge centered on keeping the platform responsive and reliable as both catalog size and transaction volume grew, which meant the data layer and the real-time pieces had to scale independently rather than as one monolith. I built back-end microservices and ETL pipelines to improve data integration across the platform, implemented real-time streaming with Apache Kafka for events that needed to propagate quickly across services, and used Redis and Elasticsearch to keep query and search performance fast as the catalog scaled. I developed secure authentication using Amazon Cognito and built React-based UI components integrated with a Python back end for dynamic, real-time data rendering on the front end. The result was a platform that scaled to support 5,000+ active users while maintaining reliable performance, with query response times cut by 92% and a 20% lift in engagement from front-end performance improvements.

Blood donation organizations were running on fragmented systems for donor management and donation tracking, which meant inefficient operations and low donor retention. The core engineering challenge was building genuine multi-tenancy, meaning each blood center needed a unified, isolated view of its own donor data and campaign performance without the platform turning into a maze of one-off customizations per client. I built the multi-tenant dashboard interfaces in React and developed reusable UI components specifically designed to hold up across each tenant's different ERP workflows rather than being rebuilt per client. I integrated the front end with the back-end APIs to keep data flow seamless across tenants, and built the donor engagement and reward system specifically to address the retention problem the organizations came in with, since centralizing the data alone wouldn't improve retention without something built to act on it. I also focused on UX for the data-heavy dashboard views, since real-time donation analytics are only useful if a coordinator can actually parse them at a glance.

Users needed a persistent AI chat experience across text and image generation models rather than a single provider wrapper, which required abstracting model differences at the back end instead of wrapping one API. I built the full-stack platform using Next.js and FastAPI, engineering real-time response streaming over authenticated WebSocket connections, including lifecycle management and automatic reconnection to survive real network drops without losing state. I designed a custom AI agent creation tool using the OpenAI Agents SDK, allowing users to generate tailored agents from a prompt and business context. I implemented JWT authentication with RS256, using cached JWKS validation and scheduled key refresh across HTTP and WebSocket routes. I built Stripe-integrated billing with subscription and credit-based plans, enforcing per-user token quotas and spending limits via real-time usage reservation, with per-request cost and latency logging for billing accuracy. I containerized the front end, back end, and an Nginx reverse proxy handling HTTPS termination and WebSocket upgrades using Docker Compose for consistent dev/prod deployment.

Traditional podcast platforms rely on basic keyword search, which makes it hard to find a specific moment or insight buried in a long episode — the search problem here wasn't matching text, it was matching meaning across audio content that had never been indexed that way before. I built the React/TypeScript front end and the APIs powering AI-driven content retrieval, implementing semantic search over vector embeddings so users could search by concept rather than exact keywords. I designed the NLP pipelines behind the summarization and key-insight extraction, working within a serverless back-end architecture chosen specifically to handle unpredictable processing loads without over-provisioning for peak demand. I also integrated analytics and event tracking throughout, since a search system like this needed real usage data, not assumptions, to tell us whether the ranking and summarization were actually surfacing what people wanted.

Dental professionals had no efficient way to track treatment plans, manage revisions, or communicate with design teams, which led to delays and miscommunication throughout the treatment lifecycle. The core problem wasn't a missing feature; it was a missing shared source of truth between two groups working asynchronously on the same case. I built responsive React components for the treatment-tracking dashboard and developed the Django back-end APIs that handle workflow management, including real-time data updates, so dentists and design teams could communicate about revisions directly on the platform rather than through scattered email threads. I integrated Stripe for secure payment processing and tuned the UI and interaction flow around how dental professionals actually move through a case, rather than using a generic dashboard layout, since the users here were clinicians first and software users second.

Businesses struggled to make sense of financial data scattered across inconsistent formats and lacked an accessible way to turn that data into a reliable forecast. I built an AI-powered financial analysis platform that accepts financial data as images, PDFs, text files, or spreadsheets, automatically extracts and structures it, then runs a custom forecasting model to surface growth projections and actionable insights. I built the Django back end and the data pipelines handling extraction and structuring, implemented the forecasting logic itself, and used Pandas for the underlying data manipulation. The result was a platform usable by technical and non-technical users alike — upload a file, get a usable forecast back, without needing to understand the pipeline underneath it.

Users in emerging markets had limited access to global fashion brands and no platform that combined social discovery with a genuine buying experience, forcing them to choose between social apps that couldn't transact and marketplaces that couldn't engage. The engineering challenge was making cross-region commerce actually work, meaning payment methods, currency handling, and fulfillment vary significantly by market, so the platform couldn't assume a single payment path would serve every user. I engineered the core front-end components in React for a seamless browsing and purchasing experience, and integrated back-end services built with Node.js and Django to handle the underlying business logic. I implemented multiple payment gateway integrations specifically to support cross-region transactions, since forcing every user through one processor would have excluded large parts of the target market. The result was a platform that let users discover, engage with, and purchase fashion products within one unified ecosystem rather than switching between separate apps.

Training centers had no centralized way to manage student enrollment, scheduling, trainer assignments, and payments, so parents, trainers, and admins had to coordinate manually across disconnected channels. The core design challenge was building a single system that worked for three very different user types at once: parents registering children and managing schedules, trainers managing their own availability, and admins needing full oversight without being buried in operational details. I built the full-stack architecture using Next.js on the front end and NestJS on the back end, with PostgreSQL and Drizzle ORM for the data layer. I implemented secure authentication combining JWT and Google Auth, integrated Stripe for training fee payments, and built the scheduling system around FullCalendar so parents could manage attendance and trainer selection without needing to understand the underlying booking logic. The admin panel gives administrators role-based access control and the ability to manage enrollments and send updates directly to parents and students.

A client needed browser-level behavior and integrations that a stock browser couldn't provide, requiring deep engine-level customization rather than something achievable through extensions or a wrapped webview. I worked on a customized Chromium build tailored to the client's specific requirements, modifying core components in C++ at the layer where interpreted languages aren't viable for the performance and control needed. I worked with Chromium's Mojo interfaces for inter-process communication, since Chromium's multi-process architecture means browser-level customization has to respect and extend that IPC boundary rather than bypass it. I built the client-facing interaction layers using JavaScript and TypeScript on top of the customized core, and used Python for the build tooling and automation that a Chromium-based project depends on, given its complex, multi-repository build system. The result was a browser tailored to the client's exact operational requirements, built on Chromium's proven engine rather than the risk and cost of building a rendering engine from scratch.