14.12 Search, Fact-Check Grounding & AI Inference Architecture

1. What It Is in Plain Words

Search and AI architecture encompasses the ingestion, vector indexing, retrieval, and inference pipelines that allow users to search across unstructured evidence and query large language models while mitigating hallucinations through factual grounding.

---

2. TruthGuard AI & Grounding Pipeline

---

3. Search Engine Technology Comparison

Search Engine	Architecture	Full-Text Search	Vector / Semantic Search	Operational Burden	Fit with TruthGuard
PostgreSQL tsvector (Chosen)	Relational B-Tree / GIN	Native SQL	pgvector Extension	Zero (Built-in)	Optimal for Incident Search
Meilisearch	Rust In-Memory	Typo-Tolerant	Experimental	Low	Good for UI Search Bar
Elasticsearch	Lucene Distributed Java	Industry Standard	Dense Vector Plugin	Very High	Overkill for Small/Med Teams
Pinecone	Cloud Vector DB	Minimal	Specialized ANN	Cloud Managed	Useful if adding RAG embeddings

---

4. Grounding & Anti-Hallucination Controls

1. Structured Entity Matching: Rather than asking an LLM whether a claim is true in free-form prose, the Edge Function extracts specific (subject, predicate, object) triples and matches them against PolitiFact and Snopes fact-check APIs.
2. Transparent Citations: The frontend renders direct URLs to authoritative investigations, providing verifiable evidence.
3. Strict Temperature Settings: Edge prompts use temperature: 0.1 to minimize creative stochastic variance.

---

5. 30-Second Spoken Interview Answer

"For search and AI inference, we use PostgreSQL's native tsvector and GIN indexing for structured incident queries, eliminating the operational complexity of running a dedicated Elasticsearch cluster.

Our AI pipeline runs on serverless Deno Edge workers, which extract verifiable claims from suspicious media and cross-reference them against PolitiFact and Snopes. By setting temperature to 0.1 and attaching direct citation links, we eliminate hallucinations and provide explainable evidence."