Haider Gillani

miQ

Founder and full-stack engineer

Every Q&A has the same pause while a microphone travels across the room. miQ removes it by turning the audience's own phones into wireless mics, with nothing to rent and no AV crew to hire. The hard part was the audio. I built a WebRTC pipeline lean enough that a thousand-person event can run over ordinary venue Wi-Fi for less than a dollar. We've run it at five events of about a hundred people each, with live transcription during the event and a summary waiting afterwards.

HP

AI Engineering Lead, HP AI

I shipped HP's first two HP AI features, Precise Scan and Scan to Email Summary, and owned them end to end as engineer and technical PM: the first prototype, the customer testing, the US launch. I also created the HP AI subscription bundle. It now reaches about a million small-business users a month, has processed more than 100 million scans, and has cut the time a scan takes in half. Underneath it sits a knowledge base of customers' scanned documents that they can ask questions of, with routing and data extraction built on Claude, TensorFlow and AWS Bedrock.

Pakistan Mission to the United Nations

Diplomat intern and AI engineer

Pakistan sits on the Security Council for 2025–26, and I spent the summer with its delegation in New York. The diplomatic work came first: following sessions and keeping track of where every country stood. Alongside it I built an agentic workflow on top of UN records that helps delegates and Permanent Representatives prepare, condense their notes and draft communications. I also built pipelines that follow each country's positions from one session to the next, on issues from the global economy to weapons of mass destruction, for Pakistan's Ministry of Foreign Affairs.

StarcAI

Co-founder, AI and full-stack engineering

The same quarterly results can read as a warning or as confidence, depending on how they're written. StarcAI rewrote public companies' investor communications so they landed the way they were meant to. I trained a six-dimension sentiment model, measuring things like investor confidence and the likelihood of investing, on more than 10,000 annotated reports and 10-K filings, and built the product in React and Next.js. A rewrite is worthless if it changes the facts, so every output was checked against its source with topic modeling and semantic similarity, and the model was tuned with RLHF to keep it honest.

NO HQ

Lead AI engineer

After the pandemic, plenty of companies were still paying for floors of offices that sat mostly empty. NO HQ rents that space out by the day to startups. The tenant earns back some of its rent, and a small team gets a proper meeting room on the few days a week it actually comes in. I built the booking platform on the MERN stack, covering more than 400 locations across the UK and Europe, and a ranking model on Airflow that matches each team to a space by price, schedule, location and size.

Universidad Abierta Interamericana

Simulation for robotics prototyping

In 2023 Argentina's inflation passed 100%, and in a country that loves its robotics competitions, teams were being priced out of the hardware. With the robotics lab at Universidad Abierta Interamericana, we built a system that models every candidate design, lets the designs compete in simulation, and leaves only the winner to be built. The lab went from eight or ten physical prototypes, thousands of dollars in parts, to a single one that cost a few hundred.

Taiwan Digital Diplomacy Association

AI/ML intern

In the first year of the war in Ukraine, Taiwan was paying close attention to how a government keeps its people's trust under pressure. I led three engineers through more than 200,000 posts from Facebook, TikTok and Twitter, gathered with Spark pipelines, and used topic modeling and sentiment analysis to see how messages from President Zelenskyy and Ukraine's foreign ministry were received at home and abroad. What we found went to Taiwan's government as guidance for speaking to its own citizens in a crisis.

Kakao Ventures

Software engineering team lead

What happens to public conversation once a product stops being niche and everyone is on it? My team studied that question through Facebook's feed, running topic modeling and sentiment analysis over 100,000 posts to find where echo chambers formed. We proposed an unsupervised genetic algorithm for recommending posts that draws people out of them, and presented it to reviewers from more than 75 countries.

Vibemap

Content recommendation

Vibemap helps cities and districts market themselves, and its app helps visitors and locals find places and events that match the mood they're in. I worked on the recommendations, using what we knew about each person to decide what they'd want to see next.