I had built a rendering engine in Rust where the fundamental primitive is a particle, not a triangle. There are no meshes. No skeletons. No polygon budgets.
Every surface is a continuous matter field held together by spring-force physics. When something takes damage it doesn't play an animation, the forces holding it together get overcome and it physically falls apart. Destruction resolution is infinite because particles don't have polygon limits.
Cloth, fluid, soft-body, volumetric light, they all come from the same particle substrate. Different spring constants, not different systems. Every other engine fakes these as separate modules. This one doesn't fake any of them.
Geometry is SDF. That means analytical normals, mathematically exact ambient occlusion from the SDF gradient, and subsurface scattering from thickness probes. No baked textures. No approximations.
Current demo runs 50 million particles in real time. 650k lines of Rust. OpenGL 3.3 renderer with 23 post-processing passes. Full visual editor with a compiled Windows .exe on the releases page.
Built this over a weekend because every C++ project I worked on needed LLM features and the options were either "wrap the Python SDK somehow" or "write the curl SSE parsing yourself again."
26 libraries across four categories — Core, Data, Ops, App. Each one is a single .hpp file. Drop in what you need, define one implementation macro, ship. No Python runtime, no package manager, no build system changes beyond linking libcurl where needed.
llm-stream has 83 unique clones already with zero promotion. Figured the rest of the suite was ready to share.
Happy to answer questions on design decisions, the single-header pattern, or anything else.
The previous documented record for simultaneous Claude Code agents is 16, set by Anthropic's own engineering team building a C compiler.
Tonight I ran 24.
What made this possible: I built a Tokio-native LLM orchestration pipeline in Rust that routes inference through a local Mistral 7B model running on an RTX 4070. The agents coordinate through governance docs (CLAUDE.md, AGENTS.md) that define module ownership and test ratios, they don't communicate directly, they communicate through the codebase.
The recursive part: the orchestrator was built using the same multi-agent workflow it enables. The MCP server that lets Claude call the pipeline natively.
Technical details:
683 tests, 0 failures
1.53:1 test-to-production ratio enforced by pre-commit hooks
9-stage deduplication pipeline with circuit breakers
Sub-millisecond pipeline overhead (inference dominates at 3.6s on Mistral 7B)
I launched a solo financial intelligence tool 2 weeks ago at $9.99/mo. Caught 6 major market moves in the first week (including a $9.9B acquisition and a +137% move). Got users in 46+ countries organically.
Then I tripled the price to $29.99 Pro / $149.99 Ultra.
Conversion behavior didn't change. If anything, traffic to the pricing page increased.
I think the problem was that $9.99 signaled "toy" to options traders who routinely risk thousands per trade. The higher price actually built trust.
Anyone else found that raising prices early improved conversion? Curious if this is specific to finance/trading products or a broader SaaS pattern.
Product: rot.trading (165K lines of Python, built solo in 9 days, $5/mo infrastructure)
Week 1 caught $14.1B in acquisitions, MASI +34%, ZIM +25%, OLB +137%, CMPS +39%, all with UTC timestamps before the moves.
165K lines of Rust, custom NLP engine (no spaCy/NLTK), GradientBoosting over 32 features, IV-aware options strategy selection, and a closed-loop suppression gate that improves win rate between retrains. Also the first financial intelligence MCP server, integrates directly into Claude as a native tool.
Full technical spec in the repo.
Every surface is a continuous matter field held together by spring-force physics. When something takes damage it doesn't play an animation, the forces holding it together get overcome and it physically falls apart. Destruction resolution is infinite because particles don't have polygon limits.
Cloth, fluid, soft-body, volumetric light, they all come from the same particle substrate. Different spring constants, not different systems. Every other engine fakes these as separate modules. This one doesn't fake any of them.
Geometry is SDF. That means analytical normals, mathematically exact ambient occlusion from the SDF gradient, and subsurface scattering from thickness probes. No baked textures. No approximations.
Current demo runs 50 million particles in real time. 650k lines of Rust. OpenGL 3.3 renderer with 23 post-processing passes. Full visual editor with a compiled Windows .exe on the releases page.
reply