Agentic AI engineering
LLM applications built end to end — multi-agent workflows in LangGraph, tool calling, MCP servers and clients, human-in-the-loop approvals, structured outputs and streaming UIs.
LangGraphMCPOpenAIClaude
AboutMe
I'm Ahmad — a full-stack developer who builds production SaaS and AI-powered products, and who likes the unglamorous middle of a project, where an idea turns into something people can actually click on.
hello
The short version
Based in
Lahore, Pakistan — working remotely, worldwide
Focus
Agentic AI systems & full-stack product engineering
Experience
4+ years across SaaS, AI products and enterprise web
Currently
Open to freelance work and full-time roles
Reach me
contact@devprogse.com
I build web applications and AI systems that feel effortless to use and are honest about what they're doing underneath.
My work sits across the whole stack. On the front I lean on React, Next.js and Tailwind to build interfaces that stay quick and understandable. Behind them I use Node, Express and MongoDB — or Python and FastAPI — to design the data and endpoints those interfaces depend on, which means I can take a feature from an empty file to something deployed without handing half of it off.
Most of my recent work is agentic AI: multi-agent workflows in LangGraph, tool calling, MCP servers so agents can safely reach real systems, RAG pipelines that answer with citations, and the evaluation and guardrails that keep all of it honest. I built a ReactFlow-based visual builder for AI agent workflows used in a commercial AI platform.
What I enjoy most is the part people rarely talk about: naming things well, handling the state where something fails, and leaving a codebase a little tidier than I found it. I also lean heavily on AI-assisted development — Claude Code, Cursor, Codex — to plan, refactor and test faster, without letting review quality slip.
20+
Completed Projects
12+
Technologies Used
100%
Remote Friendly
4+
Years Experience
how i can help:
RAG & retrieval systems
Grounded answers with citations: document ingestion, chunking strategies, hybrid vector and BM25 search, reranking, and evaluation so quality is measured rather than assumed.
LangChainpgvectorQdrantRagas
Full-stack product delivery
Production features on the MERN stack and Next.js with Python/FastAPI services behind them — API design, data modelling, RBAC, and the deployment pipeline that ships it.
Next.jsNode.jsFastAPIMongoDB
the toolkit:
Languages
HTMLCSSSCSSJavaScriptTypeScriptPythonRustC++JavaSQL
Full Stack
MongoDBExpressReactNode.jsNext.jsAngularFastAPIReduxZustandRxJSTailwind CSSShadcn UIReactFlowThree.jsReact Three FiberGSAPREST APIsGraphQLWebSockets / Socket.io
Agentic AI & LLM
LangGraphMCPOpenAI Agents SDKClaude Agent SDKTool CallingMulti-Agent OrchestrationVercel AI SDKZodPydanticPrompt Engineering
RAG & Retrieval
LangChainLlamaIndexRAG PipelinesEmbeddingspgvectorPineconeQdrantAtlas Vector Search
LLMOps & Evaluation
OpenAIAnthropic ClaudeGeminiOllamaHugging FaceEvals (Ragas, Promptfoo)LangfuseLangSmithGuardrails & PII Redaction
AI-Assisted Development
Claude CodeCursorGitHub CopilotCodex
Architecture & System Design
MicroservicesEvent-Driven (Redis Streams, BullMQ)KafkaMulti-Tenant SaaSRBAC / ABACTurborepoOpenTelemetry
Data, Cloud & DevOps
PostgreSQLRedisSupabaseDockerKubernetesGitHub ActionsAWSVercelJWT / OAuth2StripeJestPlaywrightGit
what i believe:
Readable beats clever
The best code is the kind the next person understands without asking me anything.
Ship, then refine
Something working in front of real users teaches more than another week of planning.
Details are the product
A loading state, an empty state, a focus ring — that's where software starts feeling considered.
Keep learning in public
Every project leaves me with one thing I didn't know how to do when I started it.
Got somethingworth building?
I'm open to freelance projects, collaborations and full-time roles. Tell me what you have in mind and I'll get back to you.
