Hireguard

HireGuard: An 11-agent AI swarm that automates hiring with bias detection, salary benchmarking, and career discovery—powered by Elasticsearch vector search and Gemini intelligence.

Comment

Built With

  • Database: Elasticsearch Serverless (Vector Search, Hybrid Search, ES|QL Statistics)
  • AI Models: Google Gemini 2.0 Flash (Reasoning Swarm), Gemini Embedding (3072-dim Vectors)
  • Swarm Architecture: 11 Specialized Multi-Step AI Agents
  • Protocols: Model Context Protocol (MCP) via JSON-RPC
  • Frontend: Next.js 14, TypeScript, Tailwind CSS, Lucide Icons
  • Backend: Python 3.11, FastAPI, Uvicorn, Pydantic v2
  • Libraries: pdfplumber (Text Extraction), elasticsearch-py, python-dotenv
  • Infrastructure: Docker, Docker Compose

Inspiration

The seeds of HireGuard were planted during a conversation with a friend who had applied to over 200 jobs without a single interview. She wasn't unqualified; she was invisible. Her resume was being swallowed by "Black Box" algorithms that favored prestigious school names over raw technical grit. We realized that while AI has the power to automate hiring, it also has the dangerous potential to automate bias. We wanted to build something different—a system where AI doesn't just filter people out, but actively works to keep the process fair, transparent, and helpful for everyone involved. HireGuard is our answer to a broken, opaque recruitment world.

What it does

HireGuard is a comprehensive, dual-sided AI talent ecosystem powered by a coordinated swarm of 11 specialized agents.

  • For Recruiters: It provides a "Glass Box" hiring pipeline. Beyond fitting scores, it features a Bias Auditor for real-time fairness verification. It automates the high-value "messy work"—executing ES|QL statistics for market salary benchmarking, coordinating complex schedules, and generating premium employment agreements.
  • For Job Seekers: It is an active career strategist. Seekers scan 88,518(in production) jobs via Vector Search to find their perfect matches. HireGuard then generates Interactive Interview Prep (custom behavioral questions) and a Dynamic Learning Roadmap (a 4-week structured curriculum) to help them bridge technical gaps and land their dream role.

How we built it

We built HireGuard on a robust foundation of Elasticsearch Serverless.

  • The Brains: We used Google Gemini 2.0 Flash to power our 11-agent swarm, enabling multi-step reasoning across screening, culture matching, interview prep, and pedagogical roadmap design.
  • The Search: We utilized Elasticsearch Vector Search (3072 dims) to enable semantic matching between resumes and job descriptions.
  • The Tools: We implemented a custom FastAPI MCP Server using the JSON-RPC protocol, allowing Kibana's AI Assistant to directly execute our specialized recruitment tools.
  • The UI: A high-fidelity Next.js 14 dashboard featuring infinite scroll, live telemetry streams, and a beautiful, high-contrast interface designed for enterprise clarity.

Challenges we ran into

One of the biggest hurdles was the sheer scale of the data—88,000+ jobs. Initially, our "Joins" between jobs and companies were slow, causing the UI to hang. We solved this by denormalizing our data model and implementing high-speed bulk indexing. We also faced the technical challenge of making a local MCP server talk to an Elasticsearch instance in the cloud, which required careful JSON-RPC mapping and tunneling to ensure a seamless "handshake" between the systems.

Accomplishments that we're proud of

We are incredibly proud of our Bias Audit Trail. Seeing the system "self-correct" in real-time when it detects a score disparity is a powerful moment. We also succeeded in creating a truly two-sided platform; usually, these tools only help the employer, but our AI Fit Discovery modal provides genuine, formatted roadmaps that help candidates grow even if they don't get the job.

What we learned

Building HireGuard taught us that the Model Context Protocol (MCP) is a game-changer for agentic workflows. It allowed us to turn raw Python logic into "superpowers" for our AI agents. We also learned how to leverage ES|QL for more than just search—using it for real-time statistical benchmarking (percentiles, market averages) added a layer of quantitative proof that LLMs can't provide alone.

What's next for Hireguard

We want to take HireGuard beyond the dashboard. Our roadmap includes:

  1. Slack & Teams Integration: Bringing the agent swarm directly into the chat apps where teams already work.
  2. Geo-Aware Sourcing: Using Elasticsearch's geo-spatial capabilities to help seekers find roles based on commute time and "Digital Nomad" friendly zones.
  3. Human-in-the-Loop Refinement: Allowing recruiters to "teach" the agents by providing feedback on specific screening decisions, creating a continuously improving fair-hiring model.

Built With

Share this project:

Updates