I'm a CS student from Shantou. Building low-latency, compliant LLM infrastructure for SEA developers (70% cheaper than OpenAI)
Hi everyone,
I'm a CS student from Shantou, China.
The project: I'm building an API gateway that provides access to Chinese LLMs (DeepSeek, Qwen, etc.) through Shantou's new state-approved cross-border data zone.
Why it matters for SEA developers:
Cost: ~70% cheaper than OpenAI (powered by local offshore wind energy)
Latency: ~32ms to Singapore via dedicated subsea cable
Compliance: Physically isolated "data bonded zone" (state-approved pilot)
Current status: Still early. I'm working on the API gateway and the technical integration.
Why I'm posting: I'd love to get feedback from builders like you:
Would you use something like this?
What would make you switch from OpenAI or other providers?
What are the red flags I should watch out for?
I'm not selling anything yet – just validating the idea and learning from this community.
📬 If you're interested in early access: I've set up a waitlist. The first 50 people will get $10 in free credits when the API launches.
Thank you for reading. Honest feedback (good or bad) is very welcome.
I would like to know if it can be accessed globally or is it restricted to a specific location for using it.
Oh Nice. Its good.
The #1 thing I'd need before trusting a new API provider
is transparent bot/request tracing — not just uptime
numbers, but proof that my requests actually went where
you say they went.
The thing developers fear most with regional providers
isn't price or latency — it's the unknown. Did my
request hit the model I expected? Was it logged somewhere
I didn't consent to? Did it silently fall back to a
different endpoint?
A simple per-request log (model used, region, latency,
response hash) accessible via dashboard goes a long way
toward building that trust. It's the difference between
"trust us" and "verify yourself."
On the cost vs uptime question: for side projects,
price is the entry point. For anything in production,
uptime and predictable behavior win every time. Your
32ms latency story is actually more compelling than
the 70% price cut — lead with the infrastructure
credibility, use the price as the closer.
Genuinely interesting project — good luck with the
gateway build.
Hey Zihan, smart angle with the Shantou data zone. Your waitlist page is functional but basic — want me to build you a proper landing page with email capture and social proof? $150, 48 hours. DM me if interested.
Interesting
Thanks! I'm still early on this. What would make something like this useful for you?
Founder question: for teams using Cursor/Claude Code/Codex plus app SDKs, when does provider switching, rate-limit visibility, and model alias drift become painful enough to centralize behind one gateway?
This is a great question, and honestly, it's where I'd like to take this project in the long run.
From what I've seen talking to other developers, the pain usually starts when a team has:
- 3+ different providers (e.g., OpenAI for some tasks, Anthropic for others)
Right now, I'm focused on the foundation – building a reliable, low-cost gateway for individual developers and small teams in SEA. But the unified gateway for enterprise teams is a logical next step.
Out of curiosity: at what scale (number of developers? monthly API spend?) did your team first feel the need for a centralized gateway? Would love to learn from your experience.
cost is one aspect but people look for capability also., how reliable is the output. if you can share some examples or comaprison you ran with openai vs your models
Just to clarify – I'm not building my own model. I'm building an API gateway that provides access to existing LLMs (DeepSeek, etc.) through Shantou's cross-border data channel.
So the output quality is determined by the underlying model, not my gateway. My value is:
- Cost: ~70% cheaper than OpenAI
- Latency: ~32ms to Singapore
That said, your point about needing real data is fair. I plan to publish:
- Cost comparison per 1M tokens
What specific data would make you trust a new API gateway provider?
Interesting idea. But developers will only switch if you prove reliability, model quality, and real benchmarks not just cheaper pricing. Start with a small MVP, show latency,cost comparisons, and get early pilot users to build trust.
This is very practical advice. You're right – trust comes from data, not claims. For an MVP, would you recommend starting with a simple latency dashboard (live pings to SG) or a side-by-side cost/latency comparison chart against OpenAI on a few common tasks? Also, how would you suggest finding those first few pilot users?
The cost barrier gets most of the attention but the workflow barrier is bigger for research use cases. Researchers (PhD students, postdocs) who start using LLMs for literature review, methodology checks, or academic writing hit a ceiling fast - not because the API is expensive, but because they don't have tested prompt structures for their specific tasks.
A literature review prompt that works for chemistry is broken for qualitative social science. A methodology critique prompt needs to know what 'threat to validity' means in your field. The researchers who get the most out of low-cost APIs are the ones who've spent time developing field-specific prompt libraries, not the ones who just switched providers.
Good timing on the price angle though - the unlock for SEA academic use is probably cost + workflow, not cost alone.
This is a brilliant angle. You're absolutely right: the real barrier for researchers isn't just API cost, it's the 'cold start' problem of not having effective prompts for their specific field.
I want to build exactly that: a small, open-source library of academic prompts for SEA researchers, designed to work with my low-cost API. It would turn my 'cheaper token' into a 'complete research assistant toolkit'.
Which academic field in Southeast Asia do you think has the most urgent need for this, but the least access to good AI tools? Chemistry? Social Sciences? Medical research? Your answer will help me pick my first niche.
The cost angle is useful, but I wouldn’t lead only with “70% cheaper than OpenAI.” Cheap LLM APIs are easy to dismiss as unstable, risky, or temporary. The stronger wedge is probably regional infrastructure: lower-latency SEA access, compliance-aware routing, and a cost base tied to real physical infrastructure instead of just reseller pricing.
For developers to switch, they’ll need trust before price. Clear uptime numbers, model availability, data handling, API compatibility, latency benchmarks, and fallback behavior will matter more than the discount itself.
One thing I’d think about early is the brand frame. If this becomes serious LLM infrastructure for SEA developers, it needs to sound more like durable API infrastructure than a student project or location-based experiment. Exirra .com would fit that direction well because it feels sharper, technical, and more infrastructure-grade.
This is incredibly valuable feedback. You're right. 'Cheaper' alone can sound like a red flag.
The real moat is exactly what you said: purpose-built infrastructure for SEA.
Lower latency (the direct cable to SG ~32ms)
Compliant routing (the 'data bonded zone')
Cost based on physical infra, not just reselling.
In the long run, uptime, reliability, and a clear data policy are what make developers switch.
I'll work on framing the message around 'reliable, regional infrastructure that happens to be 70% cheaper' , rather than the other way around.
A quick question for you: As a developer in this space, which data point would you trust more on a landing page?
'70% cheaper than OpenAI'
'32.7ms avg latency to Singapore' + 'Powered by state-approved cross-border data infrastructure'
Thanks again for pushing me to think bigger.
I’d trust the latency/infrastructure proof more.
“70% cheaper than OpenAI” gets attention, but it also creates doubt. Developers may wonder what tradeoff they are accepting.
“32.7ms avg latency to Singapore” plus “cross-border data infrastructure” feels more credible because it says why the cost advantage exists.
So I’d frame it like:
“Regional LLM infrastructure for SEA developers, with Singapore latency and compliance-aware routing.”
Then use the 70% cheaper point as supporting proof, not the headline.
The bigger question is trust. If a developer is moving API traffic, they need to believe the infra will stay reliable, compliant, and stable.
That is also why I’d be careful with the brand early. If this becomes serious LLM infrastructure, the name has to carry trust before people even read the benchmarks.
Exirra.com still feels closer to that infrastructure-grade direction than a name that sounds local, experimental, or cost-first.
Thanks for being so generous with your strategic advice. I'm taking it all seriously.
On the name: I hear you. For now, while I'm still in the 'student with a prototype' phase, I'll stick with TokenShantou because it's authentic to my story. But when I'm ready to launch a real product, I'll absolutely rebrand to something more neutral and infrastructure-level. Exirra.com is on my shortlist.
To your earlier point about sequencing: If you had to choose between building a simple latency dashboard first vs. getting a basic gateway working so people can actually try it, which would you prioritize? I want to prove the 'infrastructure' claim, but I also need something people can touch.
I’d prioritize the basic gateway first.
A latency dashboard proves a claim, but a working gateway proves trust. Developers believe infrastructure when they can send traffic through it, see the response, feel the latency, and understand the failure behavior.
So I’d get the gateway live first, then attach proof around it: latency, uptime, fallback behavior, routing explanation, and model availability.
On the name, I’d be more careful though.
If Exirra.com is already on your shortlist for the real product, I would not wait until after the gateway gets public traction to secure it.
That is exactly when the name becomes harder to change and the .com becomes more important. Once developers see the API, share it, bookmark docs, or build against it, the brand starts getting attached to the infrastructure.
TokenShantou works for the student/prototype story.
But if the serious product is meant to become neutral SEA LLM infrastructure, Exirra is the kind of name you’d want locked before the public gateway, not after.
Happy to discuss privately if you want to think through that transition cleanly. My LinkedIn is here:
https://www.linkedin.com/in/aryan-y-0163b0278/
For those curious about the technical side: the API will be OpenAI-compatible, so switching over is just changing the base URL. Main challenge right now is making the gateway rock-solid for production use.
What’s the #1 feature you’d need to see before trusting a new API provider like this?
Personally, I'm betting on stability and latency being the top priority for SEA devs. The subsea cable gives us a real advantage there. But I'd love to hear if I'm wrong — does a dirt-cheap price matter more than rock-solid uptime for your side projects?
Good luck! and don't forget to make your progress live at indie hackers :) for feedback
Also Check out my tool [Pterocos] at Products DB its called Pterocos : ) upvote to support me
Thanks for the suggestion! Just checked out Pterocos — interesting tool. Good luck with it!
And yes, planning to post regular updates here as I build. Appreciate the support!