diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..0780ba0 --- /dev/null +++ b/.gitignore @@ -0,0 +1,4 @@ +*.mp3 +*.bin +.DS_Store +*.wav \ No newline at end of file diff --git a/.meetings/config b/.meetings/config new file mode 100644 index 0000000..0969215 --- /dev/null +++ b/.meetings/config @@ -0,0 +1,4 @@ +STT_ENGINE=whisper +OLLAMA_MODEL=llama3.1:8b +THREADS=4 +STT_MODEL=/Users/jamestwose/Coding/ledex-demo/.meetings/models/ggml-small.en.bin diff --git a/README.md b/README.md index f96394f..f47132b 100644 --- a/README.md +++ b/README.md @@ -1 +1,8 @@ -## The local AI demo for LEDEX Events \ No newline at end of file +## The local AI demo for LEDEX Events + +Tools used: +- [pi](https://pi.dev) + +## Demos +- Sentiment analysis of reviews + learning points +- Meeting recording + summarization + action points diff --git a/local-ai-presentation/.gitignore b/local-ai-presentation/.gitignore new file mode 100644 index 0000000..5a86d2a --- /dev/null +++ b/local-ai-presentation/.gitignore @@ -0,0 +1,8 @@ +node_modules/ +build/ +dist/ +coverage/ +.DS_Store +*.log +.env* +!.env.example diff --git a/local-ai-presentation/README.md b/local-ai-presentation/README.md new file mode 100644 index 0000000..d90329a --- /dev/null +++ b/local-ai-presentation/README.md @@ -0,0 +1 @@ +## Local AI presentation for LEDEX Portimao 2026 diff --git a/local-ai-presentation/index.html b/local-ai-presentation/index.html new file mode 100644 index 0000000..a051b63 --- /dev/null +++ b/local-ai-presentation/index.html @@ -0,0 +1,13 @@ + + +
+ + ++ INTRODUCTION +
+ ++ Unlock Secure, Zero-Cost Local AI for Algarve Entrepreneurs, Nomads & Content Creators +
+ ++ Relying on big-tech online platforms invites data leakage risks, builds high monthly software subscriptions, and depends on having perfect Internet. Tap each risk card to unlock your secure local remedy. +
+ ++ {selectedParadox === idx ? card.alternative : card.detail} +
++ Running secure local AI is straightforward and divides into three simple, clear levels. Click any layer below to see how friendly desktop apps link directly to your laptop's power. +
+ ++ {layerDetails[activeStackLayer]} +
++ {layer.desc} +
++ Analogy: 4K Cinema Raw vs. 1080p Stream. Quantization compresses the model parameters (e.g., 16-bit floats to 4-bit integers), cutting RAM requirements by up to 75% while retaining ~95%+ cognitive fidelity. +
+ ++ By compressing heavy decimal parameters to integer coordinates, we reduce the speed bottlenecks caused by bandwidth throughput limiters, yielding 3x faster local response feedback directly on simple laptops. +
++ Transform unstructured data into actionable insights. Select a pipeline type and watch local AI extract, categorize, and score your business data in milliseconds. +
+ +RUNNING LOCAL INFERENCE...
+Extracting patterns โข Scoring โข Categorizing
+{item.sender}
+{item.subject}
+"{item.content}"
+{item.company}
+{item.contact}
+PIPELINE STANDBY
+Select a pipeline type above and click "RUN PIPELINE" to process data locally.
++ Map local AI targets against your company assets. Standard office equipment already houses substantial computational power. Utilize our selector to plan your deployment. +
+ ++ {hardwareDetails[selectedHardwareRow]} +
++ {goalVerdicts[selectedGoal]} +
++ Hardware: {row.hardware} +
++ {step.num}. + +
+ ))} ++ Buying or repurposing hardware is simple once you isolate your goals. Modern large models scale entirely based on **Memory Bandwidth** rather than CPU cores. Select your options below to see your optimal setup plan: +
+ ++ {spec.verdict} +
++ Download & initialize {spec.setupTools}. +
+ + {spec.purchaseTip} + ++ SaaS applications levy a lifetime licensing tax. Leverage your existing compute hardware assets to entirely sever monthly billing spikes and dynamically scale model invocations with absolute zero marginal cost. +
+ ++ Witness absolute independence in motion. Flip the sovereign kill-switch to isolate your computing workspace and verify local cognitive offline processing in high-fidelity. +
+ ++ Initiate off-grid mode to sever connection mapping hooks and activate offline model simulations on your local VRAM nodes. +
+ ++ {log} +
+ ))} + {demoStep === 1 && ( +{demoResponse}
+ )}
+ Language code (default: en, auto for whisper)
+ --threads Number of threads for STT
+ --output Output directory (default: .)
+
+Environment variables:
+ MEETINGS_DIR Config & models directory (default: ~/.meetings)
+ MEETINGS_STT STT engine (whisper/parakeet)
+ MEETINGS_STT_MODEL Path to model file
+ MEETINGS_LLM Ollama model name (default: llama3.1:8b)
+ MEETINGS_THREADS Thread count (default: 4)
+ MEETINGS_LANG Language code (default: en)
+ MEETINGS_OUTPUT Output directory (default: .)
+
+Examples:
+ # First time โ install everything
+ ./meetings setup
+
+ # Process a meeting recording
+ ./meetings recording.mp3
+
+ # Use a different LLM model
+ ./meetings meeting.wav --llm llama3.1:8b
+
+ # Use parakeet with multilingual model
+ ./meetings call.wav --stt parakeet --lang auto
+
+ # Specify output directory
+ ./meetings interview.m4a --output ./reports
+HELP
+ ;;
+ *)
+ cmd_process "$@"
+ ;;
+esac
\ No newline at end of file