mirror of
https://github.com/trustgraph-ai/trustgraph.git
synced 2026-07-01 09:29:38 +02:00
feat: fix RAG pipelines, Beep Graph branding, PWA, and ambient glow UI
Pipeline fixes: - Fix agent getting empty response from graph-rag by combining answer + explain data in single message (RequestResponse returns first msg) - Fix Doc RAG pipeline: add content field to Qdrant doc payload, seed 10 document chunks, fix type mismatches across base/flow/client - Forward explainability events from agent's KnowledgeQuery to client - Add "agent" to TERM_BEARING_RESPONSE_SERVICES for triple translation - Fix embeddings env var (OLLAMA_URL), user/collection threading, edge scoring threshold, and various protocol mismatches Branding: - Rename TrustGraph → Beep Graph (title, sidebar, settings, about) - Custom lambda + ThugLife pixel glasses SVG logo component - Forest green color palette (brand-50 through brand-900) - SVG favicon + PNG icons (16/32/180/192/512) - PWA manifest with service worker for offline shell caching - Splash screen with animated logo pulse on app load - Ambient glow background with drifting green radial blobs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
elpresidank
parent
87f6e5eb05
commit
ee45cb4850
42 changed files with 1690 additions and 153 deletions
|
|
@ -334,9 +334,87 @@ function buildTriples(): RawTriple[] {
|
|||
|
||||
entity("AI Safety", "is a focus area of", "Anthropic");
|
||||
entity("AI Safety", "is researched by", "Google DeepMind");
|
||||
entity("AI Safety", "is researched by", "OpenAI");
|
||||
literal("AI Safety", "concerns include", "alignment, misuse, and existential risk");
|
||||
literal("AI Safety", "approaches include", "RLHF, Constitutional AI, and interpretability research");
|
||||
literal("AI Safety", "approaches include", "RLHF, Constitutional AI, interpretability, and red-teaming");
|
||||
entity("AI Safety", "advocate includes", "Geoffrey Hinton");
|
||||
entity("AI Safety", "advocate includes", "Dario Amodei");
|
||||
entity("AI Safety", "advocate includes", "Ilya Sutskever");
|
||||
|
||||
// Anthropic's AI Safety approach (detailed)
|
||||
entity("Anthropic", "practices", "AI Safety");
|
||||
entity("Anthropic", "developed technique", "Constitutional AI");
|
||||
entity("Anthropic", "developed technique", "Interpretability Research");
|
||||
entity("Anthropic", "developed technique", "Red Teaming");
|
||||
entity("Anthropic", "published", "Responsible Scaling Policy");
|
||||
literal("Anthropic", "AI safety approach is", "training AI to be helpful, harmless, and honest through Constitutional AI and RLHF");
|
||||
literal("Anthropic", "conducts", "mechanistic interpretability research to understand neural network internals");
|
||||
entity("Dario Amodei", "advocates for", "AI Safety");
|
||||
literal("Dario Amodei", "approach to AI safety is", "responsible scaling with clear capability thresholds and safety evaluations");
|
||||
entity("Daniela Amodei", "advocates for", "AI Safety");
|
||||
literal("Daniela Amodei", "focuses on", "building safety-focused organizational culture at Anthropic");
|
||||
|
||||
// OpenAI's AI Safety approach (detailed)
|
||||
entity("OpenAI", "practices", "AI Safety");
|
||||
entity("OpenAI", "developed technique", "RLHF");
|
||||
entity("OpenAI", "developed technique", "Red Teaming");
|
||||
entity("OpenAI", "developed technique", "Iterative Deployment");
|
||||
entity("OpenAI", "established", "Preparedness Framework");
|
||||
literal("OpenAI", "AI safety approach is", "iterative deployment with extensive red-teaming and RLHF alignment");
|
||||
literal("OpenAI", "conducts", "external red-team evaluations before major model releases");
|
||||
entity("Sam Altman", "advocates for", "AI Safety");
|
||||
literal("Sam Altman", "approach to AI safety is", "gradual deployment to learn from real-world feedback while maintaining safety guardrails");
|
||||
entity("Ilya Sutskever", "advocated for", "AI Safety");
|
||||
literal("Ilya Sutskever", "left OpenAI to found", "Safe Superintelligence Inc focused entirely on safe superintelligence");
|
||||
|
||||
// DeepMind's AI Safety approach
|
||||
entity("Google DeepMind", "practices", "AI Safety");
|
||||
entity("Google DeepMind", "developed technique", "Scalable Oversight");
|
||||
entity("Google DeepMind", "developed technique", "Reward Modeling");
|
||||
literal("Google DeepMind", "AI safety approach is", "formal verification, reward modeling, and scalable oversight techniques");
|
||||
entity("Demis Hassabis", "advocates for", "AI Safety");
|
||||
literal("Demis Hassabis", "approach to AI safety is", "ensuring AI systems are robustly beneficial through scientific rigor");
|
||||
|
||||
// Safety techniques (detailed)
|
||||
literal("Constitutional AI", "works by", "having AI critique and revise its own outputs according to a set of constitutional principles");
|
||||
literal("Constitutional AI", "advantage is", "reducing reliance on human feedback while maintaining alignment");
|
||||
literal("Constitutional AI", "was introduced in", "2022 by Anthropic researchers");
|
||||
|
||||
literal("RLHF", "works by", "collecting human preference data, training a reward model, and optimizing the language model via reinforcement learning");
|
||||
literal("RLHF", "limitation is", "scalability of human feedback collection and reward hacking");
|
||||
literal("RLHF", "was pioneered by", "OpenAI and used in ChatGPT, InstructGPT");
|
||||
|
||||
literal("Interpretability Research", "is a", "field studying how neural networks represent and process information internally");
|
||||
entity("Interpretability Research", "is led by", "Anthropic");
|
||||
literal("Interpretability Research", "uses techniques like", "sparse autoencoders, activation patching, and circuit analysis");
|
||||
literal("Interpretability Research", "goal is", "understanding AI decision-making to detect and prevent harmful behaviors");
|
||||
|
||||
literal("Red Teaming", "is a", "security practice of adversarially testing AI systems to find vulnerabilities and harmful outputs");
|
||||
entity("Red Teaming", "is used by", "OpenAI");
|
||||
entity("Red Teaming", "is used by", "Anthropic");
|
||||
entity("Red Teaming", "is used by", "Google DeepMind");
|
||||
literal("Red Teaming", "involves", "external experts attempting to elicit harmful, biased, or dangerous responses");
|
||||
|
||||
literal("Iterative Deployment", "is a", "strategy of gradually releasing AI systems to learn from real-world use");
|
||||
entity("Iterative Deployment", "is practiced by", "OpenAI");
|
||||
literal("Iterative Deployment", "advantage is", "building societal understanding and adaptation alongside AI capabilities");
|
||||
|
||||
literal("Scalable Oversight", "is a", "research area focused on maintaining human oversight as AI systems become more capable");
|
||||
entity("Scalable Oversight", "is researched by", "Google DeepMind");
|
||||
literal("Scalable Oversight", "includes techniques like", "debate, recursive reward modeling, and amplification");
|
||||
|
||||
literal("Responsible Scaling Policy", "is a", "framework published by Anthropic for scaling AI capabilities safely");
|
||||
literal("Responsible Scaling Policy", "defines", "AI Safety Levels (ASLs) with capability thresholds and required safeguards");
|
||||
entity("Responsible Scaling Policy", "was published by", "Anthropic");
|
||||
|
||||
literal("Preparedness Framework", "is a", "framework published by OpenAI for tracking and mitigating catastrophic risks");
|
||||
literal("Preparedness Framework", "evaluates risks in", "cybersecurity, biological threats, persuasion, and model autonomy");
|
||||
entity("Preparedness Framework", "was published by", "OpenAI");
|
||||
|
||||
entity("Safe Superintelligence Inc", "was founded by", "Ilya Sutskever");
|
||||
literal("Safe Superintelligence Inc", "is a", "company focused solely on building safe superintelligent AI");
|
||||
literal("Safe Superintelligence Inc", "was founded in", "2024");
|
||||
literal("Safe Superintelligence Inc", "approach is", "pursuing safety and capabilities in tandem, insulated from commercial pressures");
|
||||
|
||||
literal("Artificial General Intelligence", "is defined as", "AI that matches or exceeds human-level intelligence across domains");
|
||||
entity("Artificial General Intelligence", "is pursued by", "OpenAI");
|
||||
|
|
@ -517,6 +595,174 @@ async function embed(texts: string[]): Promise<number[][]> {
|
|||
return data.embeddings;
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Document chunks for Doc RAG
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
const DOCUMENT_CHUNKS: Array<{ id: string; content: string }> = [
|
||||
{
|
||||
id: "chunk-constitutional-ai-1",
|
||||
content:
|
||||
"Constitutional AI (CAI) is an AI alignment technique developed by Anthropic in 2022. " +
|
||||
"It works by having AI systems critique and revise their own outputs according to a set of " +
|
||||
"constitutional principles, reducing the need for human feedback labeling. The technique " +
|
||||
"uses a two-phase approach: first, the AI generates and self-critiques responses using " +
|
||||
"constitutional principles; second, it trains on the revised outputs using reinforcement " +
|
||||
"learning from AI feedback (RLAIF) rather than human feedback.",
|
||||
},
|
||||
{
|
||||
id: "chunk-constitutional-ai-2",
|
||||
content:
|
||||
"The key advantage of Constitutional AI is that it reduces reliance on human feedback while " +
|
||||
"maintaining alignment with human values. The constitutional principles can include rules " +
|
||||
"about helpfulness, harmlessness, and honesty. Anthropic published the Constitutional AI " +
|
||||
"paper to demonstrate that AI systems can be made safer through self-supervision guided " +
|
||||
"by explicit principles, rather than requiring massive amounts of human feedback data.",
|
||||
},
|
||||
{
|
||||
id: "chunk-rlhf-1",
|
||||
content:
|
||||
"Reinforcement Learning from Human Feedback (RLHF) is a technique for training AI models " +
|
||||
"to follow human preferences. It was pioneered by OpenAI and used to train models like " +
|
||||
"ChatGPT and InstructGPT. The process involves three steps: first, a language model is " +
|
||||
"pre-trained on a large corpus; second, human evaluators rank model outputs to create a " +
|
||||
"reward model; third, the language model is fine-tuned using reinforcement learning with " +
|
||||
"the reward model providing the training signal.",
|
||||
},
|
||||
{
|
||||
id: "chunk-transformer-1",
|
||||
content:
|
||||
"The Transformer architecture was introduced in the 2017 paper 'Attention Is All You Need' " +
|
||||
"by researchers at Google Brain. It revolutionized natural language processing by replacing " +
|
||||
"recurrent neural networks with self-attention mechanisms, enabling much more efficient " +
|
||||
"parallel processing. Key innovations include multi-head attention, positional encoding, " +
|
||||
"and the encoder-decoder structure. The Transformer forms the foundation of modern LLMs " +
|
||||
"including GPT, Claude, Gemini, and LLaMA.",
|
||||
},
|
||||
{
|
||||
id: "chunk-openai-1",
|
||||
content:
|
||||
"OpenAI was founded in December 2015 as a non-profit AI research lab by Sam Altman, " +
|
||||
"Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, and John Schulman. " +
|
||||
"The organization was created with the mission of ensuring that artificial general " +
|
||||
"intelligence benefits all of humanity. In 2019, OpenAI transitioned to a 'capped-profit' " +
|
||||
"model to attract the capital needed for large-scale AI research. OpenAI is headquartered " +
|
||||
"in San Francisco and is best known for developing the GPT series of language models.",
|
||||
},
|
||||
{
|
||||
id: "chunk-anthropic-1",
|
||||
content:
|
||||
"Anthropic was founded in 2021 by Dario Amodei and Daniela Amodei, along with several " +
|
||||
"former OpenAI researchers. The company focuses on AI safety research and develops the " +
|
||||
"Claude family of large language models. Anthropic is headquartered in San Francisco " +
|
||||
"and has raised significant funding from investors including Google and Spark Capital. " +
|
||||
"The company's research focuses on interpretability, Constitutional AI, and developing " +
|
||||
"methods to make AI systems more reliable and aligned with human values.",
|
||||
},
|
||||
{
|
||||
id: "chunk-ai-safety-1",
|
||||
content:
|
||||
"AI safety encompasses research and practices aimed at ensuring artificial intelligence " +
|
||||
"systems operate as intended without causing unintended harm. Key areas include alignment " +
|
||||
"(ensuring AI goals match human values), interpretability (understanding how AI makes " +
|
||||
"decisions), robustness (maintaining performance under distribution shift), and red " +
|
||||
"teaming (adversarial testing to find vulnerabilities). Organizations like Anthropic, " +
|
||||
"OpenAI, Google DeepMind, and the Center for AI Safety are major contributors to " +
|
||||
"AI safety research.",
|
||||
},
|
||||
{
|
||||
id: "chunk-gpu-ai-1",
|
||||
content:
|
||||
"NVIDIA's A100 and H100 GPUs are the dominant hardware for AI training and inference. " +
|
||||
"The A100, based on the Ampere architecture, delivers up to 312 TFLOPS of FP16 " +
|
||||
"performance. The H100, based on the Hopper architecture released in 2022, offers " +
|
||||
"roughly 3x the AI training performance of the A100. These GPUs are used by major " +
|
||||
"AI labs including OpenAI, Anthropic, Google DeepMind, and Meta AI for training " +
|
||||
"large language models and other AI systems.",
|
||||
},
|
||||
{
|
||||
id: "chunk-deepmind-1",
|
||||
content:
|
||||
"Google DeepMind was formed in April 2023 by merging Google Brain and DeepMind. " +
|
||||
"The original DeepMind was founded in 2010 by Demis Hassabis, Shane Legg, and " +
|
||||
"Mustafa Suleyman, and was acquired by Google in 2014. Notable achievements include " +
|
||||
"AlphaGo (defeating the world Go champion), AlphaFold (predicting protein structures), " +
|
||||
"and the Gemini family of multimodal AI models. Demis Hassabis was awarded the 2024 " +
|
||||
"Nobel Prize in Chemistry for the AlphaFold work.",
|
||||
},
|
||||
{
|
||||
id: "chunk-llama-1",
|
||||
content:
|
||||
"LLaMA (Large Language Model Meta AI) is Meta's family of open-source large language " +
|
||||
"models. LLaMA 2 was released in July 2023 and made available for both research and " +
|
||||
"commercial use. The open-source approach allows researchers and developers to fine-tune " +
|
||||
"and deploy the models for their own applications. LLaMA models have been widely adopted " +
|
||||
"by the AI community and have spawned numerous derivative models and applications.",
|
||||
},
|
||||
];
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Qdrant seeding (document embeddings)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
async function seedDocumentChunks(): Promise<void> {
|
||||
// Embed all chunk content
|
||||
const BATCH_SIZE = 32;
|
||||
const allVectors: number[][] = [];
|
||||
const texts = DOCUMENT_CHUNKS.map((c) => c.content);
|
||||
|
||||
for (let i = 0; i < texts.length; i += BATCH_SIZE) {
|
||||
const batch = texts.slice(i, i + BATCH_SIZE);
|
||||
const vecs = await embed(batch);
|
||||
allVectors.push(...vecs);
|
||||
process.stdout.write(
|
||||
`\r Embedding doc chunks: ${Math.min(i + BATCH_SIZE, texts.length)}/${texts.length}`,
|
||||
);
|
||||
}
|
||||
console.log();
|
||||
|
||||
const dim = allVectors[0].length;
|
||||
const collectionName = `d_${USER}_${COLLECTION}_${dim}`;
|
||||
|
||||
// Create collection if needed
|
||||
const existsRes = await fetch(`${QDRANT_URL}/collections/${collectionName}`);
|
||||
if (!existsRes.ok) {
|
||||
await fetch(`${QDRANT_URL}/collections/${collectionName}`, {
|
||||
method: "PUT",
|
||||
headers: { "Content-Type": "application/json" },
|
||||
body: JSON.stringify({
|
||||
vectors: { size: dim, distance: "Cosine" },
|
||||
}),
|
||||
});
|
||||
console.log(` Created Qdrant collection: ${collectionName} (dim=${dim})`);
|
||||
} else {
|
||||
console.log(` Qdrant collection exists: ${collectionName}`);
|
||||
}
|
||||
|
||||
// Upsert all chunks with content in payload
|
||||
const points = DOCUMENT_CHUNKS.map((chunk, i) => ({
|
||||
id: crypto.randomUUID(),
|
||||
vector: allVectors[i],
|
||||
payload: {
|
||||
chunk_id: chunk.id,
|
||||
content: chunk.content,
|
||||
},
|
||||
}));
|
||||
|
||||
const res = await fetch(`${QDRANT_URL}/collections/${collectionName}/points`, {
|
||||
method: "PUT",
|
||||
headers: { "Content-Type": "application/json" },
|
||||
body: JSON.stringify({ points }),
|
||||
});
|
||||
|
||||
if (!res.ok) {
|
||||
const body = await res.text();
|
||||
throw new Error(`Qdrant doc upsert failed: ${body}`);
|
||||
}
|
||||
|
||||
console.log(` Qdrant: ${points.length} document chunk embeddings stored in ${collectionName}`);
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Qdrant seeding (graph embeddings)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
|
@ -756,6 +1002,10 @@ async function main(): Promise<void> {
|
|||
console.log("── Seeding Qdrant (entity embeddings) ──");
|
||||
await seedQdrant(entities);
|
||||
console.log();
|
||||
|
||||
console.log("── Seeding Qdrant (document chunk embeddings) ──");
|
||||
await seedDocumentChunks();
|
||||
console.log();
|
||||
} else if (hasQdrant) {
|
||||
console.log("⚠ Skipping Qdrant embeddings (Ollama not available for embedding generation)\n");
|
||||
} else {
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue