FAQ
AI engineering and cloud
17 questions and answers about AI engineering and cloud, from Muhammad Shahid, Senior AI Software Engineer and Backend Developer.
AI engineering and cloud
1. Do you build LLM and GenAI systems for production?
Yes. I build LLM-based features and agentic workflows that run in production, including at Joni (joni.ai), an AI product platform where I build the backend services, agent workflows and integrations.
2. Do you build RAG systems with embeddings?
Yes. I have built production RAG pipelines (retrieval-augmented generation) using embeddings and vector search: splitting content into chunks, creating embeddings, storing them for vector search, and retrieving the right context for the LLM, so its answers are grounded in your own data.
3. What is an agentic workflow, and do you build them?
An agentic workflow is an AI system that plans and carries out several steps itself: calling tools, reading data and deciding what to do next. I build them with LangGraph and n8n, keeping state between steps and passing results to the rest of the system.
4. How do you evaluate and monitor LLM features?
I trace prompts, responses and tool calls, evaluate changes against test sets before release, and monitor answer quality, latency and cost in production. It is the same discipline I learned in QA, applied to AI.
5. Do you use Python?
Yes. I use Python for AI work and scripting, alongside JavaScript and TypeScript for backends and web apps.
6. Do you use TypeScript?
Yes. I write TypeScript and JavaScript for Node.js backends and web front ends.
7. Do you work with AWS and Docker?
Yes. I containerise services with Docker and deploy and run them on AWS, alongside Heroku, VPS servers and Netlify.
8. Do you set up CI/CD pipelines?
Yes. I set up pipelines that build, test and deploy automatically, so every change is checked before it reaches production. This site, for example, deploys on every push and then submits its sitemap to IndexNow.
9. Do you design microservices and event-driven systems?
Yes. I split backends into services that talk through APIs, events and queues, and I have worked with data streaming. For Tomi I built event-driven services that index on-chain events and keep off-chain state in sync.
10. Are you a fit for a Senior AI Software Engineer role?
I combine production backend work at scale (DOP with 300M+ users, CV Maker with 1M+), hands-on GenAI engineering (LLMs, RAG, agentic workflows and evaluation), cloud and DevOps with AWS, Docker and CI/CD, and a QA background. I am open to remote and hybrid roles.
11. Do you work with LangChain or LlamaIndex?
Yes. I build with LangChain and LlamaIndex for loaders, retrievers and chains, and with LangGraph, which I am certified in, for stateful AI agents.
12. Can you integrate OpenAI and Anthropic APIs in production?
Yes. I call the OpenAI and Anthropic (Claude) APIs from production backends with retries, timeouts, streaming, rate-limit handling and usage logging, so the AI features stay reliable and their cost stays visible.
13. Do you optimise LLM cost and latency?
Yes. I cache repeated answers, use smaller models where quality allows, trim prompts and context, stream responses and batch background work, and I track cost and latency per feature.
14. Have you built multi-agent systems?
Yes. I have built multi-agent systems where specialised AI agents split a task, pass results to each other and hand off work, coordinated with LangGraph.
15. Do you do prompt engineering?
Yes. I design prompts with clear instructions and structured outputs, keep them versioned, and test every change against evaluation sets before it ships.
16. Do you use LLM tool calling and function calling?
Yes. I let LLMs call APIs and tools through defined schemas, and validate the arguments before anything runs, so agents can act on real systems safely.
17. Have you fine-tuned LLMs?
Yes. I have fine-tuned models, including with LoRA, for cases where prompting and RAG are not enough, and I evaluate the fine-tuned model against the base model before using it.