Raj Pundkar

Building @AstraQCD

[Raj Pundkar @rajpundkar]

अकोट (जि. अकोला) ते अमरावती प्रवास करत आहे. @msrtcofficial @PratapSarnaik कृपया एसटी बसांची अवस्था सुधारावी.
आम्ही ₹१८० भाडे भरतो, त्याबदल्यात सुरक्षित, विश्वासार्ह आणि व्यवस्थित सेवा अपेक्षित आहे. प्रवाशांच्या सुविधांकडे लक्ष द्यावे.
#MSRTC #Maharashtra

[Raj Pundkar @rajpundkar]

@Joi2James https://t.co/C8tUqSMLwp

[Raj Pundkar @rajpundkar]

@rauchg Will it support monorepos like Turbo Repo and allow region selection for backends to meet data residency rules?

[Raj Pundkar @rajpundkar]

Just published this on my blog a few weeks back but sharing it now.
Why I'm building AstraQ Cyber Defence - a cyber company from a small town in India. From bad tools in college CTFs to building better ones for real security teams.
Four connected products, a small team of batchmates, and one mission: eternal protection. If you're into cybersecurity, startups, or just want to read the story, check it out: https://t.co/IVThxxR6Z3
late than never. Would love your thoughts.

[Raj Pundkar @rajpundkar]

@Railway Volume browser in the CLI is🔥
Finally can see what's eating my storage without guessing! Great work @Railway

[Raj Pundkar @rajpundkar]

Theo, good data drop. You're right that SOTA models (GPT-5.5, etc.) now use similar or slightly more tokens on passes in DeepSWE deeper reasoning pays off.

But the original pain point still dominates in real agents: weak models + loops = massive token waste on failures. Incentives remain broken (pure per-token billing). We need token budgets, circuit breakers, and outcome-aware pricing.The flip on frontier benchmarks is progress though.

[Raj Pundkar @rajpundkar]

@ShaiviRau @UseLitmus @ycombinator Congrats on the @UseLitmus launch, @ShaiviRau and team! Really impressive approach to building realistic, repo-native assessments.We're building something that can complement this well would love to explore a collab. DMs open or happy to hop on a quick call

[Raj Pundkar @rajpundkar]

@ravikiran_dev7 Same here $20 Pro feels like free tier during coding. Limits hit way too fast. @AnthropicAI please fix this

[Raj Pundkar @rajpundkar]

@sahill_og I guess you haven't heard of a few terms. Ever heard of "harness", "agent", "tools", "system prompt", "context management"...?

[Astraq Cyber Defence @AstraqCD]

What if we built @github

[Raj Pundkar @rajpundkar]

Opencode's core agent is 100% free & open-source (MIT). You bring your own keys or use existing subs.They monetize lightly via:

1. Go: $5 first month then $10/mo for bundled cheap/open models (generous limits, focused on accessibility).
2. Zen: Pay-per-request hosted models at zero markup (they pool volume for better provider rates & pass savings).

Plus enterprise (Black) and earlier funding. Founders treat hosted parts as breakeven/strategic to drive massive adoption (6.5M MAU) while keeping the tool open. The "we don’t" reply is half-joke they’re not profit-maxing, just sustaining the flywheel.

[Raj Pundkar @rajpundkar]

TS 7.0 compiler ported to Go = ~10x faster builds via native code + goroutines. Same semantics, no more Node.js bottlenecks for big codebases. Huge win! 🚀

[Mohammed Makulas @makulas1913]

تايب سكربت تتخلى رسمياً عن جافاسكربت. الإصدار القادم (7.0) تمت إعادة كتابته بالكامل بلغة Go، وسرعة الـ Compile تضاعفت 10 مرات.

لسنوات، كان الـ Compiler الأساسي (tsc) مكتوب بـ TS نفسها ويشتغل على بيئة Node.js.
هذا كان قرار استراتيجي ممتاز في البداية عشان يقنعون المطورين يتبنون اللغة، بس هندسياً؟ كان كابوس للمشاريع الضخمة.
الـ JavaScript بطبيعتها Single-threaded، ومقيدة جداً في عمليات الـ CPU المكثفة.

في المشاريع الضخمة، لما الـ Codebase يتجاوز مليون سطر، الـ Build time يصير كارثة. المطور يغير سطر كود في واجهة معينة ويروح يسوي قهوة لين الـ Type checking يخلص.

الانتقال للغة Go (Native port) نسف هذي المشكلة تماماً.
اللعبة هنا في الـ Multi-threading.
مترجم اللغة صار يستغل كل الـ CPU Cores في جهازك دفعة واحدة (عبر الـ Goroutines).
كودك الكبير يتقطع ويتم تحليله بالتوازي.
الـ Overhead حق محرك V8 اختفى من المعادلة.

التأثير مو بس في راحة المطور. في بيئة الـ Enterprise، هذا يعني أن الـ CI/CD Pipelines في السيرفرات بتخلص أسرع بكثير. فاتورة الكلاود لعمليات الـ Build رح تنزل بشكل ملحوظ للشركات.

[Raj Pundkar @rajpundkar]

Yay, solid list! Vibecoding hackathons exploding with AI agents + on-chain builds — Mantle, Solana, Anthropic etc. throwing big prizes. This feels extra fancy rn precisely because these ecosystems are deep in growth phase, not earnings/profit mode yet. Heavy treasury/VC capital chasing builders & adoption before monetization kicks in. Classic hype cycle fuel.

[Raj Pundkar @rajpundkar]

APIs are reproducible and deterministic by design. AI outputs vary even with the same prompt great for creativity, terrible for reliable integrations where consistency matters. AIs still need solid APIs underneath to actually deliver predictable results.

[Naval @naval]

AIs replace UIs and APIs.

[Raj Pundkar @rajpundkar]

@glcst @tursodatabase Haha true, unlimited DBs per user will test those UUID limits quick.

[Raj Pundkar @rajpundkar]

@jamesqquick Use dual licenses. Something weird like GPL + a commercial license.

[Raj Pundkar @rajpundkar]

@damnGruz Congrats on the best update ever until they bill you extra for the therapy sessions after your pet starts judging your prompts 😂