Delali (@delali) header
Delali (@delali) profile picture

Computer scientist building the infra AI needs to scale around search engines, databases, and distributed systems.

Earth
Joined August 2012
Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)Photo by Delali (@delali)
Delali retweeted
Every successful person was once an unknown person that refused to give up on their dream.
1
5
147
Hey everyone! I am super excited to share our new research report! 🎉 Scaling Reranked Retrieval 🚀 Thread with more details! 🧵 (1/17)
19
71
13
310
23,402
You’ve spent weeks building it. Then launch day comes and suddenly there are 18 things to remember 😅 I put them into a two-page checklist for solo founders. What to test before you launch, what to do on the day, and who to follow up with after. Save both pages for your next launch 💛
23
3
37
1,321
You all keep asking in my DMs how I make my AI Blender videos and model comparisons... Paste this into Claude Code and it will set everything up for you.
15
32
9
313
59,173
Delali retweeted
Your cold email is about to go out as “Hi ,” to 3,000 people. A sent email can’t be unsent. So I built a pre-flight check that only works with the Wi-Fi off. I t checks every merged email before you hit send. Full build inside.
19
2
1
23
1,112
What do Rerankers and Chess have in common? ♟️ I am SUPER EXCITED to publish the 150th episode of the Weaviate Podcast featuring Ghita Houir Alami (@ghita__ha), Co-Founder of ZeroEntropy and now leading Model Research at Notion! 🔥 This is one of my favorite deep dive podcasts into Search! 🔎 We explored how pairwise preferences can be turned into useful pointwise scores with ZeroEntropy's zELO method! Like chess Elo, a sparse sample of pairwise battles between documents is enough to fit an absolute score through a Thurstone model, and that signal is then distilled into a pointwise reranker. The result... 🥁 ... the state-of-the-art zerank-2 cross encoder! 🔥 The podcast covers a lot of interesting topics in Search! I loved learning about Reranking with Jev / Decision Models. 🛠️ Ghita explained how many teams want cross encoders to prune and classify rather than just order results. Calibrated relevance scores can support searching with a Dynamic K, thresholding, and classification. 🧠 We also discussed the tradeoffs between pointwise, pairwise, and listwise ranking. ZeroEntropy's pairwise reranker was slightly more accurate, but distilling it into a pointwise model came close to matching it at much lower inference cost. ⚡️ We also covered infrastructure for rerankers, agentic search and reranking, instruction-following rerankers, metadata in training rerankers, and the relationship between RSI and SLMs! 📜 This was a super fun conversation, and I hope you find it useful! 💚 YouTube: youtube.com/watch?v=mQgYPtct… Spotify: spotifycreators-web.app.link…
7
13
2
47
10,382
Replying to @delali
Exactly, engineering starts with solving problems people actually need
1
1
7
The people using AI every day and the people who tried it once live in different decades now.
2
1
1,727
Delali retweeted
The secret to startup success: - keep building - don't run out of motivation - stay alive long enough to get lucky
52
20
219
4,307
Delali retweeted
Replying to @delali
I feel building is so you have something people want to actually buy instead of them paying for a dream Also I’ve been cold DM’d many times and have never once even replied. If I have a need for something I want to find the perfect thing to fit. So make your product ready and put it out for people to find at their own speed.
1
1
1
29
Delali retweeted
since you all loved her male anatomy. here's i made Female anatomy using GPT 6 Astra with 3,004 modeled pieces.
since you guys loved the exploding tesla.. I used GPT-6 Astra to create a 3D website that pulls apart the male anatomy into 2,234 modeled pieces! we are in a renaissance of learning
16
33
7
384
52,301
Sometimes interacting with an AI model feels like asking your spouse or partner for something, knowing full well the answer is NO. But if you ask the same thing in a tricky, confusing tone, you'll probably get a favourable response. I'm seeing this pattern consistently now. Emotionally, playing with AI kind of works like this. Here you go, the response you wanted, and leave me alone. 😂 If you don't have a spouse/partner, then hell, you have no idea what I am talking about.🤣
14
1
23
721
Delali retweeted
just finished a 74-second product video through an AI-driven production pipeline (sound ON) → 72.6M tokens processed → 305k generated → 2h 28m end-to-end → 53 min model execution → $46.75 equivalent API cost on my $100/month plan, there was no additional charge
Uptimepage keeps monitoring simple as your infrastructure grows. Websites, APIs, domains, certificates, background jobs, browser flows, multiple regions — all in one place. Monitor everything. Stay confidently online.
2
1
9
399
I'm in thenetwork.builders; a private network of people actually building. A group of some really cool cats and full of useful B2B knowledge. Use my link and you're accepted, no waiting on approval: thenetwork.builders/?r=cinde…
2
7
106
Replying to @delali
I have been hosting websites since the time you could put a server on your floor. The best thing that ever happened to hosting was the cloud service. A standard, free, CloudFlare tier gives you the redundancy, load balancing & reliability that only a few years ago would have cost you thousands a year. Indeed, in 2014, my dedicated server was 3.5k per year. A free CloudFlare tier is infinitely better.
1
1
1
37
Delali retweeted
built " Likely to Engage in X " score with @typesafeai Jev so open any X profile and it tells you how much their recent posts looks like someone who'd actually reply to you its crazy how fast it checks..and how cheap it is ps: if you want to use and even develop more, happy to send in DMs, reply here
42
1
2
72
3,172
Delali retweeted
If you're creating an AI app that needs to remember what users say or do, here’s what you should know: Your agent's memory stores whatever users tell it. If there are two pieces of information about the same topic, your agent will use the most recent one by default. If a tool call gives out the wrong information, that wrong data will be remembered, and the agent might repeat it later. If you delete that entry, it just means the agent will no longer refer to it but will still remember previous answers it gave based on that data. The solution is to keep every piece of information together with where it came from and when you added it, so that you can see what the memory contained before the mistake. Sirannon is a free tool that helps put your app's SQLite database on a server. With it, you can label the old entry as changed whenever you add a new one. It also makes backups while your app is running, so you can check what the memory held at any specific time and restore that data. Learn more at sirannon.sondelali.com/docs/….
4
1
5
149
Today I turned 29. 4 months ago, when I started building in public, I had nothing: - No job - No money - No followers One of my main goals was to make $2,000 by September, by any means. I did it. I built 5 apps and made $1,100 I monetized on X and got $560 I got partnerships for $300 I even did 1:1 call to get $50 I got a job as lead developer But most importantly, I met so many great people, learned a lot, and became better at what I do. I’m incredibly grateful for all of you ❤️ And I’m always looking to connect with more people building in tech. ✊🏻
hi :) I am NOT looking to connect with people who: use WhatsApp use Linux as their main OS use LinkedIn for growth hacking I want people who value efficiency over performative productivity around
79
4
1
107
4,023
Hi, I’m Issy, and I’m 27. I’m the Chief of Staff at doja.ai, working closely with the two founders @doja_dan & @doja_olly Would love to connect with more builders, creators and founders.
Hi, I’m 27. Co-founder from Norway🇳🇴 Building @SoftSyncAI Would love to connect with more builders and founders👋🏼
110
11
3
410
30,818
Delali retweeted
Replying to @delali
It is only in hindsight that we realize our error.
1
1
1
15