SLATE recruiting assistant

Slate shifts recruiting from filtering to selecting. A 3-agent AI panel (Recruiter, HM, Tech Lead) uses Gemini Flash to identify who’s worth hiring, not just who survived the filter

Comment

Slate

A slate isn't a list of applicants. It's a list of people worth hiring.

Most recruiting tools help you filter faster. Slate helps you choose better.

The Problem

Recruiting has a selection problem, not a filtering problem.

When criteria are vague, recruiters default to elimination — cutting obvious mismatches instead of confidently identifying the right person. The result: false positives slip through, hiring managers push back, and the shortlist reflects who survived the process rather than who's actually best for the role.

Traditional ATS tools match keywords, not capability. And aligning a recruiter, hiring manager, and tech lead on the same candidate means three separate reviews and a meeting nobody has time for.

The Solution

Slate gives one recruiter the perspective of three experts, grounded in criteria they define and control.

No new workflow. No black-box scores. AI assists the decision — it doesn't make it.

Key Features

1. JD Refinement & Criteria Extraction Upload any job description. Slate rewrites it and extracts three evaluation dimensions the recruiter confirms before any resume is touched:

  • Must-Meet — Hard requirements. Absence = rejection.
  • Nice-to-Have — Signals that predict excellence, not just qualification.
  • Red Flags — Patterns that predict failure in this specific role.

2. Three-Agent Resume Analysis Every resume is evaluated by three specialized agents simultaneously:

Agent Focus
Recruiter Focuses on strategic alignment, career trajectory, and overall experience level
Hiring Manager Evaluates organizational fit, soft skills, communication clarity, and long-term stability
Tech Lead Scrutinizes technical depth, identifies logical gaps in project descriptions, and assesses hard-skill proficiency

Each agent cites specific quotes from the resume with Meets / Unclear / Gap judgments anchored to confirmed criteria.

3. Tier Ranking, Not Scores Candidates are placed into Strong Yes / Yes / Maybe / No. Deliberate choice — numeric scores create false precision. The difference between 74 and 76 is noise. Tiers aren't.

4. On-Demand Deep Analysis Bulk tier assignment runs lightweight. Full quote-level reasoning, conflict detection between agents, and AI-generated interview questions only trigger when a recruiter clicks in. Fast where it needs to be fast. Deep where it needs to be deep.

Benefits

  • Filter → Select: shifts recruiter mindset from elimination to active choice
  • Zero learning curve: same workflow, AI added in — nothing restructured
  • Multi-perspective alignment without the meeting: recruiter sees all three lenses before the shortlist goes out
  • Fewer false positives: semantic evaluation catches candidates who look good on paper but can't do the job
  • Fewer manual steps: raw JD to annotated shortlist with one human checkpoint

Architecture

Pipeline

Upload JD
→ LLM refines JD + extracts criteria
→ Recruiter confirms/edits
→ Upload resumes (PDF → text via pdf.js)
→ 3 agents assign tier per candidate
→ Recruiter clicks candidate
→ 3 agents run quote-level analysis
→ Summarizer agent: consensus + conflicts + interview questions

Stack

Layer Technology
Frontend Next.js, Tailwind CSS
Orchestration LangChain
LLM Gemini 3.0 Flash
PDF Parsing pdf.js (client-side)

Why Gemini 3.0 Flash? Slate runs 4 LLM calls per candidate. Cost compounds. Flash is the most affordable Tier-1 model with enough reasoning depth for semantic resume evaluation and strict JSON output— no Pro-tier capability needed, no Pro-tier price paid.

Environment

Create .env.local from .env.example and set the Gemini API key before running the app:

cp .env.example .env.local

GOOGLE_API_KEY=replace-with-your-google-ai-api-key

Restart npm run dev after changing .env.local; Next.js reads server environment variables at startup.

Production Access Control

Production deployments require Basic Auth before any app or API route is accessible:

BIBI_APP_USERNAME=slate
BIBI_APP_PASSWORD=replace-with-a-long-random-password

If BIBI_APP_PASSWORD is missing or shorter than 16 characters in production, the app fails closed with 503.

Built at [BIBI] · Presented at Google

Built With

Share this project:

Updates