Improving Your Talent Search?

Explore This Solution
Free consultation 
•
No obligation
•
Expect a response within 24 hours

Finding 77% of Strong Candidates with Explainable AI Search

Industries
Healthcare Communications, Specialist Freelance Marketplaces, Recruitment & Staffing, Expert Networks
Services
AI/ML, Data Engineering, DevOps, Cloud, Product Design (UX/UI), Custom Software Development
Tools We used
BGE-reranker-v2-m3, BGE-large-en-v1.5, PubMedBERT, Qwen2.5-3B-Instruct, Qwen2.5-0.5B-Instruct, BM25, Reciprocal Rank Fusion, AWS, NVIDIA T4, NVIDIA L4, React, Python, FastAPI
77.4%
Strong candidates found
81.8%
Match quality
20×
Faster search

Case Study Summary

Executive summary

A specialist freelance marketplace serving the healthcare communications industry connects agencies with vetted freelancers. However, its keyword-based search struggled to identify the most relevant specialists, as expertise depends on therapy areas, scientific knowledge, deliverables and previous project experience, not just job titles and matching terms.

WhizzBridge built a domain-specific AI search and ranking system that understands natural-language requests, ranks candidates and explains each match using verifiable profile information. In a blind test before launch, it found 77.4% of strong candidates, compared with 47.2% previously, and achieved 81.8% judged match quality. A separate audit confirmed that all 416 claims checked in candidate explanations were supported by freelancer profiles.

77.4%

Strong candidates found

Up from 47.2%

81.8%

Match quality

Up from 72.4%

20×

Faster search

For a typical request

Challenges

  • Ranking specialist expertise: Search needed to evaluate therapy areas, scientific knowledge, deliverables and project experience rather than rely on keyword overlap or years of experience.
  • Understanding recruiter language: Different phrasing, abbreviations and rare technical terms required both semantic understanding and exact keyword matching.
  • Building recruiter trust: Every candidate explanation needed to be transparent and supported by information in the freelancer's actual profile.
  • Maintaining privacy and speed: Candidate data could not go to external AI services, while profile updates needed to appear in search results almost immediately without per-query AI fees.

The marketplace's keyword-based search made it difficult for recruiters to distinguish genuine specialist matches from profiles that simply contained similar terms.

WhizzBridge’s Solution

WhizzBridge developed a specialist AI search system combining semantic retrieval, keyword matching, an in-house fine-tuned ranking model and evidence-based explanations.

  • Fine-tuned candidate ranking: Adapted the ranking model over three rounds, using hard negatives in the final round to distinguish genuine matches from misleading look-alikes.
  • Hybrid search: Combined semantic search and BM25 keyword retrieval to understand natural-language requests while capturing rare exact terms.
  • Faster request understanding: Built an in-house distilled tagging model that identifies roles, therapy areas and tools in approximately 10 milliseconds.
  • Explainable results: Generated candidate explanations from verified profile facts, removing unsupported claims and identifying where each requirement's evidence came from.
  • Private, scalable infrastructure: Deployed models on dedicated AWS GPU servers, avoiding external AI services and per-query fees.
  • Near-real-time indexing: Connected profile changes to automatic re-indexing, making updated profiles searchable in approximately 19 seconds.
View UI/UX Casestudy Here

Results We Achieved

The new system helps recruiters find more relevant specialists, understand why they match and build shortlists faster.

  • 77.4% recall of strong candidates: Found approximately three in four strong candidates, compared with 47.2% using the previous search pipeline.
  • 81.8% judged match quality: Improved on the previous pipeline's 72.4% score. NDCG@10 also increased from 0.666 to 0.812.
  • Approximately 20× faster search: Reduced search time for a typical request compared with the previous pipeline.
  • Zero outright benchmark failures: Recorded no outright failures across 20 test requests, compared with four for the previous system.
  • 416 of 416 claims verified: A separate audit of 118 freelancer profiles found every checked claim across 170 candidate cards supported by profile information.
  • Approximately 19-second updates: Profile edits become searchable through automatic re-indexing without a nightly rebuild.

Recruiters now have a faster, more transparent way to discover specialist talent, with candidate data kept away from external AI services.

REACH US

Seeking a Partner? Our expert team is here for you

To get started, we would like to gather more information about your needs. We will evaluate your application and set up a free estimation call

Get Started

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.