You run recruiting at AEP Hawaii. Your network is your product — hundreds of professionals across tech, healthcare, leadership, and operations, built through years of relationships on the islands. The problem isn't finding people. It's finding the right person, right now, from data that lives in three different systems and doesn't speak the same language.
You've tried keyword search. It misses the candidate whose resume says "built single-page applications with JavaScript frameworks" when you're looking for "React developer." You've tried scrolling. It doesn't scale past a hundred profiles. What you need is a system that understands what you mean, not just what you type.
Your Thursday Morning
A client calls. They need a senior engineer with startup experience, based in Honolulu, comfortable leading a small team. They need candidates by end of week. You used to open a spreadsheet, scan LinkedIn, cross-reference notes from three different sources, and hope your memory catches the right person. Today you pull out your phone and say one sentence.
Find me senior engineers in Honolulu with startup experience and Python skills.
Deep Agent12 candidates matched across your network. Top 3:
1. Sarah K. — Score: 94. Senior Software Engineer at a Honolulu health-tech startup. 5 years Python, 3 years team lead. Founded a dev tools side project. GitHub: 47 repos, active contributor. Located: Honolulu. Available: open to opportunities.
2. Marcus T. — Score: 88. CTO at a Maui-based SaaS company (12 employees). Full-stack Python/React. Previously co-founded an ed-tech startup. GitHub: 31 repos, strong commit history. Located: Maui, open to Honolulu relocation.
3. James L. — Score: 82. Engineering Manager, enterprise background but led a startup accelerator cohort. Python, Go, JavaScript. GitHub: 19 repos, mostly private. Located: Honolulu. Last active: 2 weeks ago.
Twelve candidates. Ranked by fit. With GitHub activity, location, and availability — before you finish your coffee. The search understood "startup experience" even though none of those resumes literally say those words. It matched on founding history, small-team leadership, and accelerator participation. That's the difference between keyword search and a system that actually understands what you're looking for.
Under the Hood
Three Signals, One Score
The search doesn't rely on one method. It combines three signals to surface candidates that any single approach would miss.
Semantic understanding handles the language gap. "Machine learning engineer" matches candidates with "data scientist" experience because the concepts are close — even if the words aren't. Pattern matching encodes what you know about your market: Hawaii location gets boosted, senior titles get weight, founding experience gets priority. GitHub activity validates technical depth that resumes understate — repos, commits, languages, and contribution patterns.
The result: when you ask for "startup CTO in Hawaii with Python experience," all three signals fire. Semantic search finds broadly relevant profiles. Pattern matching boosts Hawaii-based founders with Python. GitHub data confirms they actually ship code. You get candidates ranked by real fit, not keyword frequency.
The Interface
Two Ways In, One Engine
Your team has different workflows. Some recruiters think in filters and lists — they want to browse, sort, drill into profiles. Others think in questions — "who do we have in healthcare leadership?" Both need the same data, the same search quality, the same results. So we built both.
Who do we have in healthcare leadership on the Big Island?
Deep Agent8 candidates in healthcare leadership, Big Island:
Top match: Dr. Rachel M. — Score: 91. VP of Operations at a rural health network, 12 years in healthcare administration. Previously directed clinical operations at Queens Medical Center. Located: Kailua-Kona. Strong community ties — board member of two local nonprofits.
Also notable: David W. — Score: 85. Healthcare IT Director, implemented EHR systems across three facilities. Located: Hilo. Lisa K. — Score: 79. Nursing administrator transitioning to executive roles. MBA from UH Manoa. Located: Waimea.
The web dashboard shows the same results as ranked cards with match scores, full profiles on click, and category filters for quick browsing. The AI interface accepts natural language and returns structured answers. Same search engine underneath. Different access patterns for different people on your team.
The Foundation
Messy Data Made Searchable
Your candidate data came from two sources — a resume parsing service and a profile enrichment platform. Different formats. Different fields. Different levels of completeness. Some candidates had GitHub profiles. Some had LinkedIn. Some had both. Some had neither. None of it was searchable in any meaningful way.
We consolidated 498 candidates into a normalized 27-field profile: name, location, current organization, title, skills, education, and contact information across multiple channels. The pipeline handled deduplication, skill categorization, and entity normalization — turning scattered multi-source data into a single, clean, queryable network.
How many candidates do we have with cloud infrastructure experience?
Deep Agent34 candidates with cloud infrastructure experience. Breakdown: 18 with AWS certifications or production experience. 11 with GCP or multi-cloud. 5 with Kubernetes/container orchestration as primary skill. 22 are currently employed but flagged as open to opportunities. 8 are in active job search. Top candidate: DevOps lead with 7 years AWS, currently at a Honolulu fintech. Score: 96.
Every question like this used to mean digging through files, cross-referencing spreadsheets, and relying on memory. Now it's a sentence.
The Build
Seventeen Days
September 5 to September 22, 2025. That's how long it took from "we need a better way to search our candidates" to a deployed, running platform with two interfaces and a hybrid search engine.
Days 1–3: data consolidation, normalization, and import from multiple sources into the 27-field schema. Days 4–8: search engine — vector indexing, pattern matching, GitHub enrichment, scoring calibration. Days 9–13: both interfaces — web dashboard and AI-native search. Days 14–17: deployment, testing, handoff.
498 candidates indexed. Two search interfaces live. One engine powering both. Your team was searching candidates by the end of that same month.
Your network is too valuable to search with keywords.
Every recruiter has candidate data. Most of it sits in spreadsheets, inboxes, and LinkedIn bookmarks — unsearchable, disconnected, depreciating. The difference between a recruiting firm that finds the right person in seconds and one that finds them in days is whether the data works for you or just sits there.
AEP Hawaii's candidate network is now a living, searchable system that gets smarter with every profile added. The data they already had — scattered across two services — became a competitive advantage in 17 days.