The challenge
Critical safety intelligence was scattered across 47 official sources, including the State Department, CDC, meteorological agencies, and security databases. Users toggled between sites and PDFs hunting for answers, then often emailed an analyst and waited.
Analysts manually curated and updated safety content for each country and client. Every new geography added operational drag that did not scale.
Travel advisories and health alerts changed weekly, so static PDFs and web pages were outdated the moment they were published.
- Before: user question, search State Dept site, check CDC, download a PDF, still incomplete, email analyst, wait, get an answer that may already be outdated.
- Manual curation was a variable cost that scaled linearly with every country added.
- Intelligence was a research chore living outside the product workflow.
Safety intelligence was scattered, users were hunting, and analysts were the bottleneck.
What we built
ShooflyAI built an in-app safety assistant on three layers of live intelligence infrastructure: a data ingestion layer, a RAG intelligence layer, and an in-app experience.
The ingestion layer automatically scrapes 47 official sources every hour with zero manual curation, hitting a 100% success rate on production sources and full 9.6 second refresh cycles at optimal concurrency. It runs production-ready with 0% ingestion errors and a sub-400MB memory footprint.
The RAG layer runs semantic search across curated safety intelligence and generates answers strictly grounded in official sources, filtered by country, region, and topic. Every answer carries source citations so users can verify guidance.
The experience is embedded directly in Vigilant's mobile app with seamless authentication, so users never leave their workflow, plus analytics that track query volume and geographic trends to surface content gaps.
- Data ingestion: hourly automated scraping of 47 sources across 16 countries, new advisories live within one hour of publication.
- RAG intelligence: semantic retrieval, no hallucinations, country and topic filtering, citations on every answer.
- In-app experience: embedded in the mobile app, single login, visible citations, usage analytics.
Manual analyst curation becomes automated infrastructure. Adding new countries is configuration, not headcount.
How we shipped it
Delivery ran in two phases so value was proven before scale. Phase 1 built and battle-tested the data foundation across functional, performance, API integration, reliability, edge case, and monitoring dimensions, earning a production-ready certification.
Phase 2 integrated the RAG backend with a vector database and language model, wired JWT authentication from the mobile app straight into the chat interface, and deployed the full system. The complete assistant was delivered in under 30 days from kickoff.
- Phase 1, Data Foundation and QA: 0% ingestion errors, minimal memory footprint, sub-10 second refresh cycles.
- Phase 2, In-App RAG Assistant: vector database, grounded generation, JWT auth from mobile to chat UI.
- Built like a platform, not a one-off feature, on a data pipeline designed to scale to 50+ sources.
The outcome
Once live, users got country-specific safety guidance 65% faster than manual search across government sites and static PDFs, keeping them engaged in-app instead of leaving to research elsewhere.
Safety analysts reclaimed 20+ hours a month previously lost to manual curation and repetitive advisory questions, shifting that time to strategic risk analysis and proactive threat monitoring.
The assistant reached 90%+ accuracy in analyst review, with every answer citing official sources such as the State Department, CDC, CIA Factbook, and local agencies, which builds trust and reduces liability.
- Variable analyst labor converted into fixed technology leverage that scales to any geography.
- Cited, verifiable answers reduce compliance risk and bad-advice liability.
- Higher engagement and product stickiness from intelligence that lives inside the workflow.
Faster intelligence access means better risk decisions and safer operations.
Why it matters at the platform level
Automated scraping plus RAG turns the cost of safety coverage from variable into fixed. Launching coverage in 10 new countries uses the same infrastructure with no proportional analyst cost increase.
The intelligence layer is hard to replicate and compounds over time: more usage improves retrieval, better answers drive engagement, and geographic expansion leverages the same backbone. That is a moat, not a feature.
The same ingestion and RAG foundation can power future products like incident triage, client-specific safety SOPs, premium intelligence tiers, and proactive risk alerts. One build, multiple monetization levers.
- Margin expansion through intelligence automation.
- Product differentiation through owned, grounded intelligence.
- A platform foundation for future intelligence products.
Own the intelligence layer, do not rent it.
From case study to operating system
The roadmap extends the same framework: add 20 more countries with zero platform rebuild, layer in client-specific protocols and proprietary alert feeds, and package the assistant as a premium tier or standalone product.
This is not just adding AI to the app. It upgrades the core intelligence layer in a way that compounds over 3 to 5 years, with the same capability able to power incident response, compliance workflows, and client-specific intelligence products.