# Wasim Jalali

> Applied AI engineer in Germany. I build production AI products end to end with AI coding agents: full-stack apps, internal tools and agent systems. Based in Kaiserslautern, Germany.

Wasim Jalali is a solo founder and applied AI engineer. He built Useful Apply (usefulapply.com), an AI agent that runs the whole job search, and Useful Brain, a grounded, citation-first RAG agent on Cloudflare Workers. He works with AI coding agents (Claude Code, Codex, Cursor) and writes about evals, model selection and shipping AI products. Each writing below is also available as plain markdown at the same URL with `index.md` appended, and all of them together at https://wasimjalali.com/llms-full.txt.

## Writings

- [From 41 hidden questions to zero: making a live Q&A filter safe to trust](https://wasimjalali.com/writings/from-41-hidden-questions-to-zero-making-a-live-qa-filter-safe-to-trust/index.md): A filter that sorts a live stream's Dari comments into a question queue used to hide real questions. With one strong model and a probability check before every hide, it hid none of 618 real questions and none across three recorded broadcasts.
- [7.8 times less CPU: a live-chat question filter at 400 comments](https://wasimjalali.com/writings/7-8-times-less-cpu-a-live-chat-question-filter-at-400-comments/index.md): A Chrome extension that sorts a live stream's comments into a question queue was slowing the broadcast tab down as the chat grew. After four AI models audited it and seven rounds of fixes, a 400-comment session takes 7.8 times less processor time.
- [From six confusing cards to none: fixing a live Q&A filter by replay](https://wasimjalali.com/writings/from-six-confusing-cards-to-none-fixing-a-live-qa-filter-by-replay/index.md): A teacher's live-stream question filter showed cards labelled "this person's 3rd question" that were neither hidden nor joined. Replaying the recorded broadcast through the real system reproduced all six, found the cause, and nine replays later the fix scored 115 of 115 with no real question hidden.
- [Wrongly hidden questions 4.0% to 1.9%: the test design chose the model](https://wasimjalali.com/writings/wrongly-hidden-questions-4-0-to-1-9-percent-the-test-design-chose-the-model/index.md): Twenty-eight AI models were scored on sorting a live stream's Dari comments into a question queue. The best combination cut wrongly hidden questions from 4.0% to 1.9% on the test, and four choices in how the test was built changed which model came out on top.
- [From 116 to 118 of 120: why I threw away the fix that scored perfectly](https://wasimjalali.com/writings/from-116-to-118-of-120-the-fix-i-threw-away/index.md): A company knowledge assistant that only quotes its documents reached 118 of 120 (98.3%) on its live test. The first fix repaired all four misses on replay and was discarded anyway, because it only worked for this set of documents.
- [From 72% to 95%: fixing an assistant that only quotes its sources](https://wasimjalali.com/writings/from-72-to-95-repairing-a-grounded-rag-agent/index.md): A company knowledge assistant that answers only in sentences quoted from its documents went from 72% to 95% correct on a 120-question test, without loosening a single scoring rule.
- [Same score, one ninth the price: choosing a knowledge assistant's model](https://wasimjalali.com/writings/picking-a-model-for-a-grounded-rag-agent/index.md): Five language models ran the same 120-question test behind a company knowledge assistant that only quotes its sources. The cheap one tied the expensive one at 95.0% correct, and two big-name models could not finish the warm-up.

## Optional

- [Landing page](https://wasimjalali.com/): Selected work, experience and contact
- [RSS feed](https://wasimjalali.com/rss.xml)
- [GitHub](https://github.com/wasimjalali)
- [LinkedIn](https://www.linkedin.com/in/wasimjalali)