My foundation is product design and UI/UX. Over nearly six years, I designed complex enterprise workflows, multi-tenant SaaS architectures, and more than 500 screens for Busyfile. Along the way, I expanded into frontend implementation, full-stack systems, machine learning, and applied AI. Today, I operate as an AI Product Engineer.
Design systems without runtime understanding decay into static prototypes. Machine learning systems without human ergonomics become unapproachable black boxes. I eliminate the boundary between what is conceived in Figma, verified in code, and executed across neural runtime models.
Rode to Khunjerab Pass, one of the world’s highest paved international border crossings, on my motorcycle. Unforgiving altitude, cold starts, and zero cell service.
Began in visual design and brand identity, mastering typography, color theory, layout grids, and communication hierarchy. Developed a strong intuition for clarity and user perception.
Shifted into software interface design, product discovery, user flows, and wireframing for web and mobile startups. Learned to translate ambiguous business goals into testable interface workflows.
Designed Busyfile, an enterprise taxation and compliance platform. Created more than 500 screens across client portals, firm admins, and internal operations, establishing reusable design systems.
Bridged design into production code using Next.js 15, React 19, TypeScript strict mode, and Tailwind CSS. Built custom REST endpoints and integrated state management for client applications.
Engineered production AI features at Techomatrix and independently: RAG pipelines with Qdrant, in-process ONNX query routing, LangGraph state machines, and LangSmith evaluation.
Unifying the loop: moving from product discovery and Figma systems through backend architecture, model routing, and production evaluation. Owning the space between idea and working intelligence.
Built and supported LLM-powered features, knowledge assistants and automation workflows that connected AI models with APIs, databases and business processes.
Founding product designer for Busyfile and related operational tools serving U.S. accounting and taxation firms. Led product design from early discovery through detailed workflows and engineering handoff.
Designed executive-facing visual communications and digital interfaces, combining structured information, brand consistency and clear presentation across web, mobile and business materials.
Worked on brand, marketing and product-design assignments for clients including fintech, healthcare and SaaS businesses.
Created brand identity, digital content and web-design assets for small businesses and startup-facing services.
Independent work spanning brand and visual design, product design and UI/UX, frontend implementation, AI prototypes and workflow automation. The practice evolved from identity design into broader digital-product and AI engineering work.
Defines useful AI behaviour, human-control boundaries, failure states and evaluation needs before selecting models or frameworks.
Creates reusable components, interaction patterns and documentation for large multi-role products, including Busyfile's 500-plus-screen ecosystem.
Nearly six years designing SaaS, fintech, healthcare and AI product experiences from discovery and information architecture through prototypes, systems and developer handoff.
Breaks complex operational requirements into role-aware journeys, navigation structures and understandable task flows.
Applies supervised learning, transfer learning, evaluation and deployment techniques in routing and medical-imaging research projects.
Implements embedding pipelines, similarity search, metadata filtering and retrieval hydration for grounded applications.
Develops structured prompting, information extraction, summarization, decision-support and conversational experiences using Groq and OpenAI-compatible APIs.
Models agent workflows as explicit state graphs with conditional edges, bounded retry, tool boundaries and deterministic safety paths.
Exports and serves compact routing models inside FastAPI to avoid a separate inference service and unnecessary LLM routing calls.
Builds grounded assistants with governed sources, deterministic ingestion, semantic retrieval, citation mapping and refusal when evidence is insufficient.
Used for transfer-learning experiments with VGG16, ResNet50V2 and DenseNet121 in CardioScan AI.
Builds typed API services for AI inference, retrieval, agent orchestration and integration boundaries.
Uses relational storage for canonical documents, application records, audit data, chat state and structured product data.
Used for FastAPI services, ML pipelines, LangGraph workflows, document processing, retrieval, evaluation and desktop automation.
Connects LLMs, databases, CMS platforms, publishing tools and third-party SaaS services into bounded workflows.
Designs inspectable workflows connecting LLMs, webhooks, databases, content systems and third-party business tools.
Builds responsive product interfaces, streamed AI chat experiences, dashboards and content-driven websites.
Implements responsive interfaces and design-token systems while preserving accessibility and product-specific visual language.
Uses strict types and shared contracts to make frontend behaviour and API integration safer and easier to maintain.
Packages and deploys model-backed services while accounting for cold starts, resource limits and secret management.
Uses serverless PostgreSQL for canonical documents, ingestion state, assistant records and evaluation data under free-tier constraints.
Uses Qdrant as a rebuildable vector index mapped back to canonical PostgreSQL chunks.
Uses hosted PostgreSQL and application services for prototypes, content memory and persistent product records.
Builds deterministic tests and evaluation gates for retrieval, routing, agent behaviour, frontend flows and safety-sensitive actions.
Traces agent and retrieval execution and supports evaluation of groundedness, citations, refusals and regression cases.
Tracks routing datasets, parameters, model metrics, artifacts and promotion decisions.
Accenture
Specialization: User Experience Design
DataCamp
Specialization: AI Engineering
Jawaan Pakistan
Specialization: Web Development
UXcel
Specialization: UX and Product Design
Atomcamp
Specialization: AI Agents and LLM Application Engineering
Ghulam Ishaq Khan Institute of Engineering Sciences and Technology (GIKI) / SkyLabs
Specialization: Applied Artificial Intelligence and Machine Learning
Karachi Institute of Economics and Technology (KIET)
Specialization: Computer Science
Factual AI systems must expose verifiable citations and deterministic guardrails. If context is missing, the model must clarify or refuse, never hallucinate from training weights.
Do not use a 500ms LLM call when a 3ms in-process ONNX classifier solves the problem. Respect latency budgets and cloud quotas with the same rigor applied to design systems.
True product craftsmanship means moving freely between user research, Figma component tokens, TypeScript frontend state, Python backend graphs, and evaluation telemetry.
Open to AI Product Engineering, applied AI, and lead product design opportunities.