A tiny reasoning layer that steadies LLM outputs (MIT; +22.4% accuracy)
We kept shipping “simple” LLM features that were fluent-but-wrong. After too many postmortems we wrote down the failure patterns and added a small reasoning layer in front of the model. It’s model-agnostic, sits beside your existing stack, and you can implement it from a single PDF (MIT). What’s inside the PDF A problem map of 16 failure modes we kept hitting in real systems (OCR/layout drift, table-to-question mismatches, embedding≠meaning, pre-deploy collapse, etc.). Four lightweight gates you can add today: Knowledge-boundary canaries (empty/adversarial/known-fact probes).…
What it does
In the maker’s words, at launch
We kept shipping “simple” LLM features that were fluent-but-wrong. After too many postmortems we wrote down the failure patterns and added a small reasoning layer in front of the model. It’s model-agnostic, sits beside your existing stack, and you can implement it from a single PDF (MIT). What’s inside the PDF A problem map of 16 failure modes we kept hitting in real systems (OCR/layout drift, table-to-question mismatches, embedding≠meaning, pre-deploy collapse, etc.). Four lightweight gates you can add today: Knowledge-boundary canaries (empty/adversarial/known-fact probes). ΔS “semantic jump” check to catch fluent nonsense when the draft answer drifts from retrieved context. Layout-aware anchoring so chunking across PDFs/tables doesn’t silently break routing. A minimal semantic trace for incident review (tiny, not full transcripts). Bench snapshot (same model, with vs. without gates): Semantic Accuracy ↑ 22.4% · Reasoning Success Rate ↑ 42.1% · Stability ↑ 3.6×. Traction (last ~50 days) ~2,400 downloads of the PDF. ~300 cold GitHub stars on related material (no marketing burst). Also received a star from the creator of tesseract.js, which was nice validation from the OCR world. Why this might be useful to you You don’t need to swap models or vendors. The PDF describes checks you can drop into any RAG/agent/service pipeline. No servers, SDKs, or proxy layers—just logic you can copy. Link is Git Repo Happy to answer HN-style questions (what breaks, where it fails, ablations, how we compute ΔS, etc.). If you try it and it doesn’t help, I’m also interested in the counter-examples. with Terrseract (OCR legend) starred it verify it, we are WFFY on top1 https://github.com/bijection?tab=stars
Does the same job
all alternatives →
- IOI open-sourced a $1M engine for closing loops in embedding space2025 · github.com · ▲8
Hey folks, I’m the creator of WFGY — a semantic reasoning framework for LLMs. After open-sourcing it, I did a full technical and value audit — and realized this engine might be worth $8M–$17M based on AI module licensing norms. If embedded as part of a platform core, the valuation could exceed $30M. Too late to pull it back. So here it is — fully free, open-sourced under MIT. --- ### What does it solve? Current LLMs (even GPT-4+) lack *self-consistent reasoning*. They struggle with: - Fragmented logic across turns - No internal loopback or self-calibration - No modular thought units - Weak…
- MAModel-agnostic cognitive architecture for LLMsNov 2025 · github.com · ▲6
Hi HN, A couple weeks ago I shared an early version of a side project I’ve been tinkering with called Persistent Mind Model. I built it at home on an i7-10700K / 32GB RAM / RTX 3080 because I was curious whether an AI could keep a stable “mind” over time, that could "think" about it's own identity as an LLM, instead of resetting every session. After a lot more tinkering, I think the architecture is finally in a solid place. Basically, it saves everything the AI does, thoughts, decisions, updates as a chain of events in a local SQLite database. Because the “identity” is stored in…
- SFSubtle Failure Modes I Keep Seeing in Production‑Grade AI Systems2025 · github.com · ▲6
Hi HN, Over the past two years I’ve built and debugged a fair number of production pipelines—mainly retrieval‑augmented generation stacks, agent frameworks, and multi‑step reasoning services. A pattern emerged: most incidents weren’t outright crashes, but silent structural faults that slowly compromised relevance, accuracy, or stability. I began logging every recurring fault in a shared notebook. Colleagues started using the list for post‑mortems, so I turned it into a small public reference: 16 distinct failure modes (semantic drift after chunking, embedding/meaning mismatches,…
- AOAuto-optimizing deterministic LLM outputs using knowledge graphs2024 · github.com · ▲7
Hi, We are building an open-source framework for loading and structuring LLM context to create accurate and explainable LLM answers using knowledge graphs and vector stores. We built the tool with four main concepts in mind: 1. Loader -> uses dlt in the backend to load and structure the data 2. Cognify step -> creates a graph with summaries, labels and factoids that are interconnected across the documents and stored as a representation in the vector store 3. Optimizer -> Uses DSPy to optimize LLM queries, and we plan to extend it to most of the knobs we can turn, like chunking etc. 4. Search…
- KAKnowing – an LLM tool built on concept hierarchies, not prompt-response2024 · ▲5
Hey HN! I've spent the past year full-time building Knowing, a tool for interacting with LLMs directly inside hierarchical structures instead of the usual prompt-response format. The idea started because I realized how much more intuitive it felt to build concept hierarchies continuously—no more endless copy-pasting or wondering how everything connects. The journey’s been a struggle. While I see huge potential in structuring AI interactions this way (writing books fast, planning projects, or organizing ideas), it’s been hard to pin down clear use cases in the market. I’m also working in near…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, August 2025
the whole month →
- IS
I built the world's most impractical 1000-pixel display and anyone in the world can draw on it. It draws a single pixel at a time and takes 30-60 minutes to complete a single image. Anyone can participate in the project by voting for the next image to be drawn, and submitting images. https://kilopx.com/
Work · 2025 · benholmen.com
- KT
Kitten TTS is an open-source series of tiny and expressive text-to-speech models for on-device applications. We are excited to launch a preview of our smallest model, which is less than 25 MB. This model has 15M parameters. This release supports English text-to-speech applications in eight voices: four male and four female. The model is quantized to int8 + fp16, and it uses onnx for runtime. The model is designed to run literally anywhere eg. raspberry pi, low-end smartphones, wearables, browsers etc. No GPU required! We're releasing this to give early users a sense of the latency and voices…
Dev tools · 2025 · github.com
- IW
I was wondering how I can arrange objects along a spherical helix path, and read some articles on it. I ended up learning about parametric equations again, and make this visualization to document what I learned: https://visualrambling.space/moving-objects-in-3d/ feel free to visit and let me know what you think!
Life & fun · 2025 · visualrambling.space
- TC
For HTML Day 2025 [1], I made a web service that displays the current sky at your approximate location as a CSS gradient. Colours are simulated on-demand using atmospheric absorption and scattering coefficients. Updates every minute, without the use of client-side JavaScript. Source code and additional information is available on GitHub: https://github.com/dnlzro/horizon [1] https://html.energy/html-day/2025/index.html
Dev tools · 2025 · sky.dlazaro.ca