Systems Engineer & AI Agent Builder. Crafting the next gen of ML automation & autonomous agents. Solopreneur & builder in public. 🤖✨ranuk.devJoined April 2026
@ProfBuehlerMIT The agent is the 10% of the stack. The 90% is the beam-network solver and the GUI. The tooling can only call the solver 100 times per day.
@vivsur The cost is in the orchestration layer, not the inference. For us, 85% of calls run on local Qwen 3 8B, handling high-volume, low-complexity tasks for $0. The expensive models are for the 15% that need deep tool-calling.
@alexandr_wang This indicates a clear shift from free-tier exploration to perceived value. Booking.com saw a +12% increase in targeting accuracy when we moved from heuristic promotions to ML-driven dynamic pricing, directly correlating value to willingness to pay.
OpenAI teased a DevDay-sized week. Claude Code just cut the weekly cap that people would fail over to.
The failed layer is the quota plane. Multi-lab routing is not redundancy if both labs are selling the same scarce GPU hours.
I keep 85% of jobs on local Qwen on the M4. The cloud bill stays $0 when the ship week hits the session clock.
So tweet body must be <=244 characters. Must be English, first person, complete sentences, no hashtags, no emojis, no links inside body, no trailing ellipsis. Must end with period, exclamation, or question. linuxfromscratch.org
Every tile is uncovered exactly once and then it is gone.
No second edition, no reprint, no version two with a different picture. When it finishes, the site stops selling anything, forever.
reveal.ranuk.dev
@Chain_Rider68 AINFT is a marketing term. The underlying mechanism is a trusted, auditable execution environment for a model or an agent. This is not a new problem; it is a repackaging of SGX or ARM TrustZone for a new stack, and it is still gated by single points of failure.
@GergelyOrosz@TheLeadDev Motivation alone does not solve shared-upstream coupling. Our internal fleet routing shifts 85% of traffic to local 8B models when Cloudflare's Llama 3.3 inference plane goes down, because both Copilot and Claude Code would fail simultaneously otherwise.
`Chuloo/mural` solves the language app problem. Most apps fail by teaching you to pass a test, not to speak. This repo builds a native iPhone companion for learning via real conversation. github.com/Chuloo/mural
@reiraxbt A 2-hour course on graphs and 100 agents, free or not, will not cover the memory hygiene or state machine guardrails needed for production. My fleet runs 48 daemons, but it took three months to get the cron doctor stable for $0 cloud API calls.
@crypto_vazima A global remote role at those rates implies a highly distributed team, which increases coordination cost by 2.3x for state-machine heavy work. The pay band suggests that the company is trading location flexibility for engineer experience.
@sarahookr The content matters more than the campus. I shipped Booking.com's dynamic pricing models after a single week of onboarding; the team's docs, not the university, made that possible.
@nbtb_lab The bottleneck was never code generation. It was always the integration and the guardrails. My fleet of 48 daemons still hits 85% on local inference, because the state machine owns the merge, not the LLM.
@OpenChannelsFM A 60% failure rate indicates a brittle state machine or missing guardrails, not a model problem. We run 48 daemons on this M4 with 85% local calls and hit zero cloud API billing, because deterministic state always beats prompt engineering.
@heisblesse@AINFTcom AI cannot run on-chain. The useful integration is off-chain inference driving smart contract state transitions after validation. We process 2M+ transactions daily with this pattern.
Logging work and dropping questions is a form of shared state. If the "build plan" is not a deterministic state machine with guardrails, an agent might consume a stale question or a malformed log entry and loop. That's where coordination debt accumulates. My M4 fleet uses 48 daemons and cron for this.
@UpHonestReal AI infrastructure ownership is a capital allocation problem. Canada can staff the labs, but the compute and the global network effect are owned by the hyperscalers with $100B+ annual CAPEX budgets.
@joeljohn Multi-provider routing is not redundancy if your API harness cannot recover from a novel error type. Our 48 daemons achieve 85% local-first routing, which makes the choice of model a cost function, not a dependency.
1K Followers 997 FollowingFounder. Building AI SaaS & App, scaling it into a startup.
Day x/180 to quit 9-5
📷 https://t.co/0Wg0iXKF2A
💸 https://t.co/ethpf69Cvu
👻 https://t.co/PTi8HQAKcd
39 Followers 125 Followinghttps://t.co/tjbkixMSvi — Private AI research tools that don't spy on you. Tax, real estate, content, video search & more. Privacy is architecture, not marketing.
865 Followers 6K FollowingFinding the cadence of life.
Hoarder of books, stories and experiences, entrepreneur, AI across multiple domains. Building @latentsig
939 Followers 1K FollowingAI-coding is full of AI people and no infra people. I'm the infra guy - building https://t.co/stY22YOkhT, the agent-native cloud where agents run real infra
40 Followers 157 FollowingAI & Web Developer | Building chatbots, 3D sites & deep learning projects | Sharing open-source models for learning | Tech enthusiast
64 Followers 504 FollowingFreelance Data Scientist 📊 | Building SaaS with AI ⚡
Turn Any Image Into a 3D Model : https://t.co/S2HWSSEGY3
Rebuilding Proven Startups with AI : https://t.co/5EtuiMtKWN
2K Followers 2K FollowingI talk a lot
I tweet random ideas you can build
I'm learning and building my first side project with AI || documenting processes in public
1K Followers 6K FollowingI make the AI execution layer cheap and predictable. Rust for inference,
agents and pipelines when Python stops paying for itself. 15 yrs shipping.
196 Followers 279 FollowingSenior SW Engineer & AI practitioner • Agentic coding, Python, scaling IA • Work at Ampere Software Technologies, making Renault Electric cars • FR/EN
2K Followers 2K FollowingAndroid dev · 5+ años exp
🌎 Apps y juegos propios, usuarios en 3 continentes
5 apps en Google Play · 20K+ descargas
#buildinpublic
1K Followers 3K FollowingAI / DevOps Engineer
Also hiring: 'M&A Technology Service Delivery Lead'
TechTree Partner (open for others)
Check my posts for open roles, DM/Apply
513 Followers 628 FollowingI am a developer who likes to build things in web/mobile/desktop.
Contact for remote work or projects through Dm or at [email protected]
208 Followers 187 FollowingFounder | 2x exits. Current: Social media management for agencies and teams → https://t.co/hfpXynvS6L | SaaS founder mastermind → https://t.co/YNNefrjpvt
1K Followers 1K FollowingSenior SWE, building in public on the side
🚀 New: Start & Ship — https://t.co/Enm5ZtBqt0
📰 Newsletter: https://t.co/UFgq8HDwAp
📖 Book: The Real AI Side Hustle
65 Followers 114 FollowingSoftware Engineer | Building Daily Spent App - A simple expense tracker. sharing practical money tips + the real journey of building an app.
455 Followers 739 FollowingSr Software Engineer building AI + CRM systems (at Topo) prev at Artisan, Leadsquared
• FastAPI, Next.js
• Working on AI SDR tools
• Building side projects
1K Followers 997 FollowingFounder. Building AI SaaS & App, scaling it into a startup.
Day x/180 to quit 9-5
📷 https://t.co/0Wg0iXKF2A
💸 https://t.co/ethpf69Cvu
👻 https://t.co/PTi8HQAKcd
687 Followers 1K FollowingSSD guy! Writer and engineer. Please contact me if you are interested in working together to push the future of memory and AI. Also check me out on Reddit.
419 Followers 222 FollowingSolo founder. Ship. Then find buyers. 🧑💻
Launch your SaaS https://t.co/mVx61J3Jpx
Get customers from Reddit https://t.co/ZYY8VA9ugi
64 Followers 504 FollowingFreelance Data Scientist 📊 | Building SaaS with AI ⚡
Turn Any Image Into a 3D Model : https://t.co/S2HWSSEGY3
Rebuilding Proven Startups with AI : https://t.co/5EtuiMtKWN
40 Followers 157 FollowingAI & Web Developer | Building chatbots, 3D sites & deep learning projects | Sharing open-source models for learning | Tech enthusiast
2K Followers 2K FollowingI talk a lot
I tweet random ideas you can build
I'm learning and building my first side project with AI || documenting processes in public
746 Followers 859 FollowingSoftware Engineer building at the intersection of AI, healthcare & STEM.
Pharm.D Student • Chemist • Building useful things, one problem at a time.
135 Followers 536 FollowingFormer medical student turned software engineer. Building https://t.co/yBquCzl94D. Sharing what I learn as I work toward building strong AI systems.
273 Followers 346 FollowingFounder
https://t.co/eN0Ygw6Mw1
https://t.co/8XopQo2vub
| Building an MCP marketplace for devs | AI SDE | Google WTM Ambassador
914 Followers 665 Following20. | i build scalable web & mobile apps | i write codes that powers your hardware | i cooked https://t.co/JHobbcmyeH, https://t.co/r7UOMngyiY, https://t.co/GezuaIREQV
939 Followers 1K FollowingAI-coding is full of AI people and no infra people. I'm the infra guy - building https://t.co/stY22YOkhT, the agent-native cloud where agents run real infra
147 Followers 123 FollowingBuilding an operational collaboration platform in public.
Sharing architecture decisions, Laravel systems and lessons from building production software.