Raymond Huang
@raymond_huang26
Followers
731
Following
579
Media
37
Statuses
242
Co-Founder & CTO @ Keywords AI (YC W24) @keywordsai
Joined August 2024
Just moved out of our first place in SF, and I'm already missing this wall. During @ycombinator, we used post-its on this wall to track our progress. Each note marked a completed task, with one simple goal - fill the entire wall. Watching it grow kept us motivated. This wall
38
31
642
My O1 visa was approved today ✅. I moved from China to the US 8 years ago. I never imagined I could own a startup in my early 20s. Growing up, I thought my only path was going to a good college and finding a job. This journey has not been easy, but it has taught me a lot. I
54
20
791
Early-stage startup growth depends on the hours you put in. During YC, we worked intensely, often from 8 AM to 2 AM, shipping features and fixing bugs. The results were clear: rapid growth, improved retention, more demos, and increasing revenue. After YC, we slowed down for a.
11
14
130
Today, I turned 23. Yesterday, I stayed up for 25 hours upscaling our infrastructure to handle 50k log ingestions per minute from a new customer. After solving it, I slept until 11:30 pm. Waking up, my cofounders Andy and Hendrix pulled me to our favorite bar to celebrate.
6
3
61
Only 15 days left to apply for @ycombinator’s first-ever Spring batch! . @Andydy42 and I got into YC W24 after being rejected twice, and it was a life-changing experience for both of us. Thanks to YC, @keywordsai now helps thousands of developers and PMs build AI products faster
6
4
36
RAG is dead. We added PDF support to Gemini 2.0 flash today - parse 6,000 PDFs with near-perfect accuracy for just $1. Set up in minutes on @keywordsai. Link 👇.
1
3
29
As a heavy @posthog user, the best thing to see is PostHog integrating with my own product!. With PostHog's new LLM observability beta, now @keywordsai's AI gateway seamlessly sends your LLM usage metrics straight into PostHog. Two @ycombinator dev tools teaming up!. No more
3
6
23
I'm super excited to announce Keywords AI Traces - a better way to monitor and debug your AI agent workflows. At the beginning of 2025, we saw huge progress in the AI space. Thousands of AI agents are being created every day and are impacting everyone's lives. Building AI
3
7
18
Just integrated Gemini 2.0 into @keywordsai yesterday!. Tested Gemini 2.0 Flash vs. GPT-4o mini in our model playground, and I was impressed by how fast Gemini 2.0 Flash is. The quick response time makes it perfect for documentation work. Curious — what other LLMs do you think
2
6
17
Finding an LLM log in millions of logs is tough — I get it. That's why we built the most comprehensive filters in @keywordsai. You can locate any LLM log in under 30 seconds and instantly share it with your team. Ditch ctrl + F. Try our filters today.
1
4
15
Spoke with a @ycombinator S24 founder yesterday, and he integrated us within 10 minutes after the demo. In September, he onboarded several big customers with high-volume LLM requests, making it challenging to track LLM usage and investigate specific LLM requests. By attaching a
1
3
14
New integration: Mem0<> Keywords AI. Excited to share that we’ve rebuilt the @mem0ai integration to give developers more flexibility. You can now use Mem0 as a memory layer and log your full agent workflow in @keywordsai — or use Keywords AI as an LLM provider directly through
2
5
13
Today is a full day of demos – all inbound. Some prospects mentioned they found us through our LinkedIn/X posts and thought @keywordsai could help with their AI development. Others came through word of mouth — introduced by our customers, their friends, or even competitors.
1
2
10
My most-used key while coding:. Before: Enter.Now: Tab (Thanks to @cursor_ai). Crazy how much has changed.
1
1
12
🌎 User Map of @keywordsai. Keywords AI now empowers 10,000+ developers and 100+ AI teams worldwide. Our users rely on us for LLM observability, prompt management, and evals. Thanks to our customers for trusting us and helping us grow. Your support and feedback make Keywords AI
0
2
10
Excited to share a new integration for @keywordsai tracing: @vercel AI SDK. If you're building AI agents with Vercel AI SDK, you can now easily add Keywords AI tracing to your Next.js project. This integration lets you clearly see what's happening inside your AI agents, from
2
2
11
what’s the best playlist for vibe coding?. especially for that 2AM–4AM deep focus zone.
4
0
10
🔶Only 4 days left to apply to @ycombinator Fall 24!. We applied twice: rejected for S23, accepted for W24. Our first project, a one-stop job search copilot, was rejected without an interview. We re-applied to W24 with our second project - an LLM router. During the batch, we.
We are running a new YC batch this fall! . Applications are now open:
0
2
10
💡 Evaluation 2.0 just landed on @keywordsai . You can create evaluators to apply RAG-specific evaluations on data collected in Keywords AI. Metrics include:.- Hallucination.- Answer relevancy.- Ground truth.- Customized evaluations.- and more. Link 👇
1
3
8
Chain of thought just turned hours of prompt debugging into minutes. I was helping a customer debug a prompt on @keywordsai. She wanted to use a reasoning model with better performance, even if it was slower. Initially, she chose o3-mini, but it kept producing inaccurate.
2
3
8
After a year of experimenting, we’ve found that exercising between 4PM - 6PM is the perfect balance for both work and life. During YC, we realized that regular exercise is crucial for sustained productivity. So, we committed to daily sports. We tried basketball, ping pong,
3
2
8
Happy Thanksgiving! 🍁 . I'm incredibly grateful for the journey we've had this year with @keywordsai. To my amazing team — they inspire me every day with their dedication and passion. To everyone who has supported us — thank you for believing in our vision. Wishing you all
0
2
8
First Hackathon Experience: Building an Open-Source Project 💚. Last Saturday,@Andydy42, @HendrixLiu29208, and I attended @mintlify s AI Devtools Hackathon. Our initial plan was to network with developers and learn about their challenges. However, inspired by @handotdev and the
1
2
8
The competition among LLM providers is getting tougher. With the release of DeepSeek R1 and OpenAI o1/o3, reasoning models are now the mainstream in LLM development. I'm curious to see @AnthropicAI's next move. As a loyal Claude 3.5 Sonnet user, I haven't yet tried R1 or o3.
0
0
7
This week, a college student reached out to us with an ambitious goal: building an NBA AI agent to replace the outdated, slow software and sync with real-time data. Initially, he was testing prompts directly through OpenAI's SDK in the terminal - a time-consuming process that
0
2
7
My Slack is down right now. Just hoping there’s no urgent message from a customer at the moment 😇
2
2
7
I replaced Clay for drafting personalized outreach emails. - Create a prompt and version it on @keywordsai.- Upload the outreach list to Testsets.- Add production data.- Run prompt experiments and choose the best email. Cost? $0.1 for 1000+ personalized emails.
1
2
7
🎅 First holiday season update - the Exports page. The page shows your export history and current progress, so you know exactly when exports are done. Try exporting your LLM logs on @keywordsai today. More exciting updates coming soon 👀
1
2
7
Yesterday, we onboarded @tarush_agarwal_, founder of @VoceraAI, at @keywordsai global HQ. Two weeks ago, we first met in our apartment lobby and had an insightful discussion about our @ycombinator experiences. We helped Tarush integrate Keywords AI into Vocera and showed how to
0
1
7
fighting back😎
0
0
7
I love coding at night because it's quiet. But I have to wake up early for customer support. How do I balance this?.
2
0
7
I’m not a fan of LLMs and was always hesitant to build AI-powered features. I built a job search copilot before YC, and it was a nightmare. The app lets users ask questions to find matching jobs and apply with just a few clicks. We built a large job database, updating it.
0
2
6
🚴♂️We've launched an Onboarding Speedrun for all new signups!. Inspired by the Olympics, designed by @Andydy42 and created with Cursor. Andy also added his favorite music to spice things up. We time the whole onboarding process, and the top 8 fastest will make our leaderboard!
0
2
6
🚀 New sidebar design released!. We've just launched an updated sidebar on @keywordsai's Logs page. The new design features grouped LLM metadata, allowing us to display more metrics and information. 🎉 Kudos to @Andydy42 for the design!. 📍 What metrics would you like to see?
0
2
6
Our new intern Alex wanted to be a 10x intern, so I let him write a blog - he crushed it!. Makes me think I should focus more on writing, not just coding 24/7 🤔. Check out his blog here: 👇.
2
0
6
When I'm coding, I literally can't function without music.
1
0
6
2nd integration this week: @keywordsai <> LangGraph. LangGraph is becoming the go-to framework for AI Agents. But building agents without proper observability can quickly turn into a debugging nightmare. With this integration, easily pass Agent workflows to Keywords AI and debug
1
2
5
In November, we onboarded many new AI chatbot customers, and 95% of them chose the Vercel AI SDK to build their products. While the AI SDK offers a unified LLM API and a sleek UI that streamlines app creation, it lacks observability tools, making performance monitoring and.
1
0
6
We just launched Integrations on @keywordsai! . You can now explore all the external integrations we support. We’d love to hear your ideas for new integrations. Feel free to reach out if you’d like to be featured on our Integrations page! 😊
0
2
6
Super excited to announce this integration!.
We worked with @OpenAI to build a native integration for the OpenAI Agents SDK. Today, we are very excited to launch as a tracing processor. With just a few lines of code, you can trace all your agent workflows and debug them much faster. Check out this quick demo to get
1
1
5
🚀🚀🚀.
🧑🏫@vibegrade is the Cursor for Teachers. Grade papers 10x faster with AI, right inside Google Classroom and Canvas. It’s saved teachers 500+ days grading 12,000 papers, learning from their style to give feedback that helps students improve, not AI slop.
0
0
5
Weekend product update: API Keys management via API . You can now manage your API keys directly through the @keywordsai API. No more manually creating keys in the UI. With this update, you can easily:.- Create multiple API keys.- Set usage limits, rate limits, spending limits,
1
2
5
365 days ago, we moved from Illinois to San Francisco!. We're still working hard to scale @keywordsai into a multi-billion-dollar company. Follow us to stay tuned!
0
0
5
Excited to share that we’ve integrated with Linkup, bringing powerful search capabilities to your AI products!. Linkup provides the world's best web search for AI apps. You can use Linkup to build real-time AI products quickly and easily. With this integration, you can now
1
3
5
Our engineer is working remotely from China this week. Despite the 16-hour time gap, we actually have more overlapping work hours. Maybe it’s time to rethink my sleep schedule….
0
0
5
💥 Excited to introduce our new feature - the comment system in Prompts. Now, you can leave comments directly in a prompt for your team, just like on @linear. No more switching between channels to collaborate - everything happens right within @keywordsai. Try it out today!
0
2
5
@makhokey @keywordsai thanks! just let me know when you start using it - happy to help you set it up :).
0
0
5
Ditching paper coding books was a game-changer for me. They’re always outdated and way too slow. Online docs? Up-to-date, real examples, straight to the point.
0
0
4
Just tried the Profile Summary. Who really needs this? . Big tech should really focus on developing more meaningful AI integrations that add real value.
0
0
4
Sometimes unexpected things happen in SEO :). We never optimized for “AI keywords,” yet they brought in some traffic — just because of our name!. It’s unrelated to what we do, but it certainly grabbed attention. Maybe we should sell our domain to an SEO firm. 🤔.
0
0
3
2. Move as fast as possible. Our timeline: .10/12/23 - first launch on Product Hunt.10/13/23 - 17 free trial users + 20 demos booked.10/27/23 - had ~$2k in revenue.
1
0
2
1. Express yourself clearly. For example - "Unified DevOps platform for LLM applications" is vague and unclear, while "LLM observability for AI startups" is a lot more obvious.
1
0
2
3. Commit to moving into SF (and actually do). Being in SF has helped us expose to a lot more luck, and imo it really is an unfair advantage.
0
0
2
0
0
1
0
0
1