AI prospect research & data enrichment
LLM Contact Discovery
A production-ready workflow that turns a plain-language university search into a reviewable, exportable directory of verified contacts.
Business problem
The challenge
Finding the right contacts across university websites is slow, inconsistent, and difficult to scale. Useful information is scattered across faculty directories, profile pages, tables, subdomains, and PDFs, while titles and departments vary widely between institutions.
The system needed to discover contacts efficiently without treating an LLM as an unchecked source of truth. Every result had to remain traceable, reviewable, and useful to an operator.
Technical solution
Our approach
A hybrid architecture pairs deterministic crawling and validation with selective AI reasoning, so the system stays scalable, inspectable, and grounded in source data.
- 01
Interpret the search
OpenAI web search translates a plain-language query into likely university candidates, with confidence and source reasoning presented before a discovery run begins.
- 02
Crawl deterministically
A resumable breadth-first crawler explores sitemaps, directories, common subdomains, profile pages, and HTML tables. Deterministic parsing does the bulk of discovery; AI fills targeted gaps.
- 03
Validate and enrich
Contacts are deduplicated, checked for valid names, sanitized through rules and batched AI review, then enriched with public emails, phone numbers, office details, and source URLs.
- 04
Keep people in control
Progress states, confidence scores, flags, source links, selective enrichment, and CSV export make the automation inspectable instead of opaque.
Results & impact
A repeatable research system.
The finished product replaces scattered manual browsing with one traceable workflow, while keeping operators in control of review and enrichment.
page crawl ceiling
A resumable queue and persisted crawl state support broad institutional sites without losing progress.
contacts per enrichment pass
Priority ordering and controlled concurrency keep enrichment focused on the highest-value records.
operator stages
Search, discovery, and results turn an open-ended research task into a clear, repeatable workflow.
Visual assets
From search to actionable results.
Product demo
See the workflow in action.
Follow the full journey from a university query to discovered, reviewable contacts.
Open on YouTubeTech stack
Built for production.
Designed and built end-to-end as a full-stack AI system: product workflow, application architecture, crawler, AI orchestration, persistence, background processing, validation, and deployment.