home.social

Search

59 results for “scottgal”

  1. Random little tool today. Was having issues with Skia.Svg in my lucidVIEW app so built a little totally .net one. This is just an AoT CLI tool github.com/scottgal/lucidview/ which will convert svg to png easily.
    Not 100% (no animation etc) but roughtly equivalent to most browsers.
    #aot #svg #cli #dotnet

  2. Finally got proper Mac support for lucidVIEW my .NET AoT Markdown Viewer.
    Little 60Mb app which is:
    Fully .NET native makrdown document rendering through Avalonia on Mac, Windows and Linux (no WebView so doesn't slurp 300Mb of your RAM ).
    Optimized for quick browsing super fast startup, async Mermaid diagram rendering
    Fully native .NET Mermaid diagram rendering ...etc, also has a few extensions through mostlylucid.naid my fork of naiad the C# Mermaid renderer
    Cross platform PDF output / printing of Markdown docs
    etc...
    All the niceties, really just the markdown downcument reader *I* wanted so it's fast, low footprint and just works...
    As usual Free & Open Source
    github.com/scottgal/lucidview/

    #markdown #dotnet #aot #mermaid

  3. v1 Of DoomSummarizer is out.
    It's a crazy deep research / auto knowledgebase system. Point it at a directory of word docs, pdf and markdown it'll index it all then answer questions about the contents. Point it at a url it'll parse the content, index it and tell you what it's about.
    Crawl your company's knowledgebase? It'll automatically become a support AI.

    Want to know what your biggest invoice was, when you sent that angry letter etc...all local, all private, all open source (unlicense) . Quick two as unlike most RAG systems it MINIMIZES token use.

    #llm #ai #rag #search #localllm #ollama #onnx github.com/scottgal/lucidrag/r

  4. Ok leaving it for the evening. Have a little json based file format for ascii art. Means I can make a tiny player so any CLI can play animations / images as ascii art. github.com/scottgal/mostlyluci
    The braille mode shown here fits a WHOLE lot of detail into a tiny space in the console!
    #asciiart #csharp #aot #terminaltools

  5. Made my little tool an mcp tool github.com/scottgal/mostlyluci

    github.com/scottgal/mostlyluci

    Warning it will DESTROY context windows. It can use 10k tokens for a small image to actually render it inline! But it has multiple modes to save files, extract gifs from video, get info etc...etc...

    #mcp #aot #csharp #imagesharp #ai #claude

  6. Resist and unsubscribe: resistandunsubscribe.com

    Scott Galloway docente ed esperto di finanza lancia una campagna di boicottaggio economico contro le BigTech e le aziende che sostengono l'ICE.

    Ecco come motiva la sua iniziativa:
    "Dobbiamo riconoscere che il presidente è del tutto indifferente all'indignazione dei cittadini, ai tribunali o ai media. Risponde a una sola cosa: il mercato. L'arma più potente per resistere all'amministrazione è uno sciopero economico nazionale mirato della durata di un mese — una campagna coordinata che colpisca le BigTech e le imprese che sostengono l'ICE — per infliggere il massimo danno con il minimo impatto sui consumatori. In sintesi, la via più breve per il cambiamento senza danneggiare i consumatori è uno sciopero economico mirato alle aziende che guidano i mercati e sostengono il nostro presidente"

    Segue un elenco di #BigTech da boicottare disdicendo gli abbonamenti ai loro servizi

    #EconomicStrike #boycott #IceOut #Resist #Unsubscribe #ScottGalloway

    @scuola
    @informapirata
    @RFancio
    @Khrys
    @lindasartini
    @quinta
    @filippodb
    @mcp
    @lgsp
    @Puntopanto
    @casilli
    @steko
    @nemobis
    @gubi
    @jenshansen
    @dado
    @DarioZanette

  7. Finalising the first StyloBot 'pack' (defining the architecture mainly).
    This one is for ASP.NET - think Grafana like charts, adaptive throttling, openapi reading, dynamic protection against exploits etc..and live updating of endpoint policies.

    Will be the starter - comes with the FOSS package but a £5/month sub gets realtime updates and more functionality (auto honeypot etc).

  8. Prediction; Claude Code will soon ship with a small local model. Seems an obvious improvement.

  9. Older version of StyloBot finally @chocolatey approved (older will get updated at the weekend)

    But FINALLY

    choco install stylobot

    WORKS!
    This weekend for Winget & a few others.

    FINGER'S CROSSED StyloBot should RTM this weekend (so mostlylucid.botdetection v7 / stylobot v1) .

    Stage 1 in my business plan.
    Make the Free, Open Source Version *so good* that it's my advertising *by just working*.

    Commercial (realtime management, fleet / postgres / sql etc db support, distributed mesh architecture) will arrive June 1st.

    LIKELY first in the guise of £5-20/mo 'packs' for the likes of ASP.NET, Magento, Wordpress (ANY version...will provide an automatic security upgrade).

  10. Two new draft StyloBot articles today:

    1. On the upcoming TypeScript SDK for StyloBot (behavourally aware UX for Node / client side)

    mostlylucid.net/blog/typescrip

    2. What makes that (and other) SDKs like Go /Caddy support possible..a SideCar mode for StyloBot - low latency low resource StyloBot running alongside any other service.

    mostlylucid.net/blog/sidecar-a

    #aot #dotnet #npm #go #stylobot

  11. OK my StyloBot hours begin (after my two contracts).
    Today is getting the new SideCar mode in shape for the draft article.
    This is what lets Node, Caddy etc...work with it in high performance modes. So .NET service running over gRPC in 1-2ms for a bot decision.

    SO also building a small SDK with Typescript (client and server) & Go (Caddy) implementations. And it has a neat 'template' mode which accepts liquid templates and renders HTML controls allowing a hybrid rendering mode.

    Makes it really useful beyond .net (so the behavioural UX idea from ASp.NET works everywhere mostlylucid.net/blog/behaviour.

    #stylobot #dotnet #caddy #frauddetection

  12. OK my StyloBot hours begin (after my two contracts).
    Today is getting the new SideCar mode in shape for the draft article.
    This is what lets Node, Caddy etc...work with it in high performance modes. So .NET service running over gRPC in 1-2ms for a bot decision.

    SO also building a small SDK with Typescript (client and server) & Go (Caddy) implementations. And it has a neat 'template' mode which accepts liquid templates and renders HTML controls allowing a hybrid rendering mode.

    Makes it really useful beyond .net (so the behavioural UX idea from ASp.NET works everywhere mostlylucid.net/blog/behaviour.

    #stylobot #dotnet #caddy #frauddetection

  13. OK my StyloBot hours begin (after my two contracts).
    Today is getting the new SideCar mode in shape for the draft article.
    This is what lets Node, Caddy etc...work with it in high performance modes. So .NET service running over gRPC in 1-2ms for a bot decision.

    SO also building a small SDK with Typescript (client and server) & Go (Caddy) implementations. And it has a neat 'template' mode which accepts liquid templates and renders HTML controls allowing a hybrid rendering mode.

    Makes it really useful beyond .net (so the behavioural UX idea from ASp.NET works everywhere mostlylucid.net/blog/behaviour.

  14. OK my StyloBot hours begin (after my two contracts).
    Today is getting the new SideCar mode in shape for the draft article.
    This is what lets Node, Caddy etc...work with it in high performance modes. So .NET service running over gRPC in 1-2ms for a bot decision.

    SO also building a small SDK with Typescript (client and server) & Go (Caddy) implementations. And it has a neat 'template' mode which accepts liquid templates and renders HTML controls allowing a hybrid rendering mode.

    Makes it really useful beyond .net (so the behavioural UX idea from ASp.NET works everywhere mostlylucid.net/blog/behaviour.

    #stylobot #dotnet #caddy #frauddetection

  15. Part 3 of the StyloBot Release Series is up.

    mostlylucid.net/blog/stylobot-

    This one is less about bots and more about the reality of long-running .NET based systems: everything that learns from traffic eventually accumulates.

    Came from one of my periodic reliability reviews where StyloBot’s vector layer had drifted to 13GB on the .NET Large Object Heap due to the wrong abstraction (in-process HNSW behaving like an unbounded cache).

    The interesting part wasn’t the fix. It was recognising that the architecture itself was wrong for the runtime pattern.

    Covers:

    how I periodically review long-running services

    using dotnet-counters, dotMemory and dotTrace to find growth

    why “just add a cap” is often the wrong answer

    replacing unbounded ANN structures with bounded hot caches + compacted persistence

    taking the vector layer from 13GB LOH to <6MB

    The broader point applies to any system that “remembers”:

    bot detection, fraud scoring, recommendations, anomaly detection, RAG pipelines, adaptive systems.

    Fix the shape, not the symptom.

    #dotnet #aspnetcore #performance #architecture #ai #rag #observability

  16. Wrote an article and published a nuget package sitting here in the hospital. mostlylucid.net/blog/simple-oc
    (RDP over tailscale rules 🤓)
    Companion CLI and nuget packages linked at the top. Makes OCR & Named Entity Extraction SUPER EASY.

    #ocrr #llm #ner

  17. Problem: we keep using frontier LLMs as glue for jobs that are already solved.

    Solution: run OCR + NER locally in C# with ONNX Runtime. Deterministic extraction on ingest. Store the entities. Use an LLM later only if you actually need synthesis.

    OCR with Tesseract, then BERT NER via ONNX in .NET. No Python, no cloud, no tokens.

    This is my 'for beginners' article. I'm DEEP in OCR but realised I never explained the quickest way to do this *locally*.

    mostlylucid.net/blog/simple-oc

    #CSharp #DotNet #ONNX #OnnxRuntime #OCR #NER #LocalAI #RAG #DocumentAI

  18. New in lucidRAG: AudioSummarizer, Reduced RAG + Constrained Fuzziness for audio.

    Compute once at ingestion: deterministic signals + auditable evidence (transcript, diarization turns, tiny speaker clips).
    Query-time: retrieve facts → optional LLM narrates. No accent/genre guessing. No speaker naming. Offline, pure .NET.

    mostlylucid.net/blog/audiosumm

    #RAG #dotnet #ONNX #Audio #Privacy #SignalProcessing #LLM

  19. DocSummarizer Part 3 is the “I went too far” deep dive: how to build *proper* document summarization as a pipeline, not “shove text into an LLM and pray”.
    mostlylucid.net/blog/docsummar

    It covers:

    * sentence embeddings (MiniLM/BGE/GTE) + mean pooling + tokenization gotchas
    * ONNX Runtime in C# to run embedding models locally (no Python, no cloud)
    * RAG done right: chunking + segment extraction + salience scoring
    * hybrid retrieval: dense + BM25 fused with RRF, plus MMR to keep summaries diverse
    * real production failure modes (tokenizer mismatch, rare terms, OCR garbage, citation hallucination) + “coverage honesty” when you’re sampling

    If you want *why it works* (and what breaks), this is the post. If you just want to run the tool, Part 2 is enough.
    mostlylucid.net/blog/docsummar

    #llm #ai #onnx #csharp #dotnet