{"slug": "using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4", "title": "Using embeddings + LLM verification to match entities across data sources — dual T4 setup, looking for what others have tried", "summary": "A developer building an entity-matching pipeline across multiple data sources is seeking community input on their dual T4 GPU setup, which uses embeddings for candidate generation and an LLM (Ollama Cloud) for verification. The pipeline scans sources on a schedule, embeds records for similarity search, and uses LLM-based binary verification to reduce false positives, with results shown on a live dashboard. The developer is tuning embedding model choice, verification reliability, throughput on dual T4s, and handling the false positive tail, and asks for others' experiences.", "body_md": "Hi all,\n\nI’ve been building a pipeline that needs to match “the same real-world thing” as it’s described differently across multiple independent data sources — different wording, different granularity, sometimes different structure entirely. Wanted to share the approach and see what other people have found works (or doesn’t).\n\nCurrent setup:\n\nScan stage — pulls candidate records from several sources on a schedule\n\nMatch stage — embed everything and do similarity search to generate candidate pairs (running on 2x T4s, splitting the embedding load across both)\n\nVerify stage — because embedding similarity alone throws too many false positives, I run a second pass through an LLM (currently) using Ollama Cloud) that looks at each candidate pair and makes a binary yes/no call with reasoning\n\nResults feed a live dashboard\n\nWhat I’m still tuning / would love input on:\n\nEmbedding model choice (x) - curious if anyone’s compared options specifically for short, noisy, informally-worded text (not clean prose)\n\nVerification pass reliability — the LLM-as-verifier step is the biggest lever for precision right now. Anyone gone deeper on structured verification prompts vs. fine-tuning a smaller classifier instead of paying LLM cost per pair?\n\nThroughput on dual T4s — is splitting embedding batches across two T4s actually the right move vs. a single bigger card? Curious what others’ cost/throughput tradeoffs looked like at similar scale\n\nFalse positive tail — the long tail of “technically similar but semantically different” pairs is where most of my remaining errors live. Anyone found a good pattern for catching that without just throwing more LLM calls at it?\n\nNot trying to be cagey about the domain, just keeping this focused on the ML/infra side. Happy to go into more detail on any part if useful.\n\nSeeking wisdom", "url": "https://wpnews.pro/news/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4", "canonical_source": "https://discuss.huggingface.co/t/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4-setup-looking-for-what-others-have-tried/178475#post_1", "published_at": "2026-08-05 15:36:16+00:00", "updated_at": "2026-08-05 15:52:38.793533+00:00", "lang": "en", "topics": ["machine-learning", "artificial-intelligence", "large-language-models", "ai-infrastructure"], "entities": ["Ollama Cloud", "T4"], "alternates": {"html": "https://wpnews.pro/news/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4", "markdown": "https://wpnews.pro/news/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4.md", "text": "https://wpnews.pro/news/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4.txt", "jsonld": "https://wpnews.pro/news/using-embeddings-llm-verification-to-match-entities-across-data-sources-dual-t4.jsonld"}}