Rajan Bajaj
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

OpenLens is a modular, extensible, AI-powered search engine. Basic search is the core. Everything advanced is an independent, configurable plugin — add, remove, enable, or disable without touching core. Original idea in one line: search you can inspect — your own crawl, your own ranking, every decision explained. How it works: • Provider independence: OpenSearch, authenticated external APIs (e.g. Bing), and a custom index built by OpenLens crawler + indexer. All return normalized results through one provider manager. Custom index is the long-term differentiator — now the Postgres-backed default, single source of truth, no external search dependency. • 10-stage deterministic pipeline: query processing → provider selection → retrieval → normalization → mandatory filtering → deduplication → scoring → reranking → enrichment → response assembly. • Plugins own capabilities: query understanding, child-safe / privacy filters, trust scoring, fairness / perspective reranking, enrichment, AI summarization. Optional failures isolate, mandatory safety fails closed. • Ranking owned by OpenLens, not providers: relevance, credibility, freshness, diversity, plus per-mode profiles for default, child-safe, news, shopping, travel. Gender-neutral, evidence-aware — never popularity over truth. • Safety by design, AI as optional: safety and privacy enforced in the tool layer, not the prompt. AI Mode answers only from the index with [n] citations, Powered by … attribution, and honest extractive fallbacks. • Values: neutrality over engagement, safety by construction, transparency, your data / your index. Non-goals: maps, web-scale crawling, third-party search APIs, tracking profiles.