Muhammad Nauman Arshad

I build AI-powered applications end to end, LLM and generative model backends and the web interfaces people use them through.

AI & LLM Engineering
Multi-agent systems and agentic workflows: orchestration, tool calling, structured outputs, validation and self-correcting retries
LLM integration across providers (Mistral, Groq, OpenAI-compatible APIs), including fallbacks, rate-limit handling and cost control
Prompt engineering and evaluation backed by measurement.
AI safety and content moderation: prompt injection defence, text and image screening
Generative AI & Computer Vision
Diffusion models with PyTorch and Hugging Face diffusers: text-to-image, image editing, inpainting, reference-guided generation
Self-hosted model inference servers: GPU memory optimisation, model caching, cost-efficient deployment
Object detection, segmentation, background removal and depth estimation
Model benchmarking and selection
Backend Development
Python, FastAPI, Pydantic
PostgreSQL, SQL, Alembic migrations
Background job queues and workers, state machines, real-time progress streaming (SSE)
REST API design, authentication and sessions
Docker and Docker Compose
Frontend Development
Next.js, React, TypeScript
State management (Zustand, TanStack Query), shadcn/ui and Tailwind
Interactive canvas editors, drag-and-drop interfaces, PDF generation
Engineering Practice
Automated testing (pytest), including LLM code tested offline
Strict typing (mypy) and linting (ruff)
Git workflows, clear technical documentation and written decision records
I take products from architecture to UI, and I’m strongest where AI meets real-world constraints: limited budgets, unreliable third-party APIs, and outputs that must be checked rather than trusted.

I’m a good fit for:
AI SaaS MVPs and chatbots/agents
Generative image and content tools
LLM integrations into existing products
FastAPI backends
Next.js web apps and dashboards