Inspiration
Currently, the process of finding the best candidates for a job position is complicated. First, HR analysts must thoroughly understand the profile being sought, then analyze each available resume and select a few to contact. Second, startups and small companies lack the financial resources and time to complete this entire process over several weeks or months. And finally, contacting each potential candidate requires certain communication skills to better approach people, rather than a standard template that doesn't connect with potential employees.
Thanks to AI and document analysis, it's possible to leverage AI not to replace an HR professional, but to enhance it by connecting the human and the artificial to improve the process.
Video demo (over 3 minutes): https://www.youtube.com/watch?v=qygakV7yWH4 Video demo (3 minutes): https://www.youtube.com/watch?v=fhK44XJ0feE
What it does
TalentPilot is an AI-powered recruitment platform that helps HR teams find the perfect candidates for their job openings. By leveraging vector embeddings and natural language processing, TalentPilot matches job descriptions with candidate profiles to identify the best fits.
TalentPilot Chat is designed for several key use cases in modern recruitment:
Rapid Candidate Sourcing: Quickly find qualified candidates by describing job requirements in natural language, eliminating the need for complex Boolean search strings.
Skills-Based Matching: Go beyond keyword matching to find candidates with relevant skills and experiences, even if they don't use identical terminology in their CVs.
Passive Candidate Engagement: Identify and reach out to passive candidates who match specific role requirements but haven't applied directly.
Talent Pipeline Building: Create talent pools for future hiring needs by describing ideal candidate profiles and storing matches for later outreach.
Role Analysis and Optimization: Analyze existing job postings against your candidate database to identify potential improvements or gaps in role requirements.
Competitive Intelligence: Understand the talent landscape for specific roles by analyzing candidate availability and qualifications.
TalentPilot Chat provides value to multiple stakeholders in the recruitment process:
- HR Recruiters: Save time by quickly finding qualified candidates without manual CV screening.
- Hiring Managers: Get immediate access to relevant candidates when defining new roles.
- Talent Acquisition Teams: Improve efficiency and quality of hire through better candidate matching.
- Startups and Small Businesses: Access enterprise-level candidate matching capabilities without dedicated recruitment teams.
- Executive Recruiters: Find high-level candidates with specific skill combinations and experience patterns.
- University Career Centers: Help students and alumni find job opportunities that match their qualifications.
How we built it
The platform consists of two parts: an API (Backend) integrated with the OpenAI SDK and TiDB Cloud, and a web app (Frontend) so that the user can interact, upload CVs, search for similar profiles, and chat with the agent to find the best profiles according to job specifications. Technically, it was built with:
- Node.js (v14 or higher)
- npm (v6 or higher)
- TiDB Cloud account: For the database and vector searching with AI embedding models.
- OpenAI API key (or any other OpenAI SDK-compatible service, like Kimi/Moonshot).
Using TiDB Cloud with Vector-Based Search, TalentPilot has the following advantages:
- Semantic Understanding: Match candidates based on meaning and context rather than exact keyword matches.
- Improved Relevance: Find candidates with relevant skills even if they describe them differently than your job posting.
- Fuzzy Matching: Handle variations in terminology, abbreviations, and phrasing across CVs and job descriptions.
- Multilingual Support: Process CVs and job descriptions in multiple languages with consistent matching quality.
- Continuous Learning: Refine matching accuracy over time as more data is processed and feedback is incorporated.
- Complex Pattern Recognition: Identify subtle patterns in experience and skills that traditional search methods would miss.
Challenges we ran into
Honestly, I didn't encounter any challenges or inconveniences during development, since using the database with TiDB Cloud is the same as using MySQL with its libraries, so the learning curve was almost zero. I only had to understand the vector aspect a little more, for storing data using embeddings, but I relied on the OpenAI SDK for support.
Accomplishments that we're proud of
Developing a multi-agent platform, since it's not just a prompt that responds to the user, but rather several connected agents (analyzing the search request in detail, searching for the best profiles, ranking them, generating recommendations, and delivering a complete summary), makes the platform well-defined and, at the same time, expandable with more features.
This project also stemmed from the experience of friends in HR who struggle with this candidate search problem, so before development, I ran a validation process for the idea to create something functional and useful from day one.
What we learned
Using TiDB Cloud, specifically the vector search function (I used VEC_COSINE_DISTANCE) to find similarities, as I have previously used AI, but not with vector databases and embeddings, has given me greater insight into what I can do.
What's next for TalentPilot
The next step is to design the following functionalities:
- Authentication of users with different roles (admin, HR, candidates, and more).
- Feedback to the model, so HR can indicate if a candidate has already been hired or if communication was ineffective.
Also, transform this product into a real, scalable service.
Log in or sign up for Devpost to join the conversation.