$ cd ~/build $ ./contact
cat now.txt

now

Where I am, and where I'm pointed.

status open to ai engineering roles
building invisiblesearch — a live search engine over GitHub
next ship an ai project with real retrieval, end to end
./explain --how-i-build

build

The patterns behind an AI product, drawn out instead of listed. Retrieval, reasoning, and the space in between.

retrieval-augmented generation cyan = retrieval · lime = generation

Documents get chunked and embedded once, up front. At query time, the same embedding model turns the question into a vector, pulls the closest chunks, and hands them to the model as context — so the answer is grounded in real text, not just what the model memorized.

docs chunk embed vector store top-k retrieve prompt assembly llm structured output
agent loop — ReAct think → act → observe
think act observe
search http db mcp
embedding space nearest-neighbor retrieval
docs code chat query
ls -la ./stack --curriculum

stack

What I've studied and used to build the two projects below, organized the way the field is actually structured. Tap a card for the full topic list.

ls -la ./projects

projects

The same fundamentals an AI product runs on: typed APIs, structured responses, caching, clean deploys.

./InvisibleSearch

live github search engine

A minimal search engine over all of GitHub, live. Queries proxy straight to the GitHub search index through a FastAPI backend — the same pattern an LLM app uses to call an external API and shape its response.

  • Mapped raw API responses into typed, paginated results with a debounced, keyboard-navigable type-ahead — structured outputs, not loose JSON.
  • Rate-limit aware by design: a 60s per-query cache, structured 429s with reset timestamps, and quota warnings surfaced in the UI.
  • Containerized with Docker Compose; pytest suite runs against a mocked GitHub client.
[python][fastapi][javascript][docker]
↗ live ↗ github

./NetZero

cross-platform expense app

A cross-platform app for tracking expenses offline. A clean exercise in state and structured data — the same discipline a retrieval layer or agent memory needs.

  • Every record persists in local storage with no connection required — offline-first by design.
  • Built around a central state layer and reusable components, with validation at the input boundary.
[react-native][expo][javascript]
↗ live ↗ github
degree BS Software Engineering — University of Central Punjab (UCP), Lahore
coursework DSA · System Design · Applied Machine Learning · Formal Methods
focus AI Engineer — applying pre-trained models via LLMs, RAG, and agents
languages English (professional) · Urdu (native)
./contact.sh

contact

Have a model worth putting to work? Let's talk.

[email] aqifarain@gmail.com [x] @aqifahmed_ [github] @aqifahmed [linkedin] /in/aqifahmed [resume] download pdf ↓