talent-ranker/submission_metadata.yaml at main · 07MADARA/talent-ranker · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
# ============================================================================
# Team identity
# ============================================================================
team_name: "JOYBOY_07"

primary_contact:
  name: "Kukkuluri Siva Kumar"
  email: "sivakukkuluri82@gmail.com"   # Fill this in
  phone: "+91-6281799856"           # Fill this in

team_members:
  - name: "Kukkuluri Siva Kumar"
    email: "sivakukkuluri82@gmail.com"
    role: "AI Engineer"

# ============================================================================
# Code and reproducibility
# ============================================================================
github_repo: "https://github.com/07MADARA/talent-ranker"

sandbox_link: "https://talent-ranker-dxmhdh6vijjjv6rwdmp99a.streamlit.app/"

reproduce_command: "python rank.py"

# ============================================================================
# AI tools declaration
# ============================================================================
ai_tools_used:
  - "Used Google Gemini for architectural brainstorming, trap-detection logic formulation, and UI/UX Streamlit design generation."

# ============================================================================
# Approach summary
# ============================================================================
methodology_summary: |
  We engineered a Multi-Stage Cascading Pipeline optimized for strict CPU latency constraints.
  Stage 1 applies deterministic guardrails to eliminate ~80 honeypots and keyword-stuffers (e.g., catching Marketing Managers with AI keywords).
  Stage 2 uses lightweight TF-IDF semantic vectorization to match candidates' actual career histories against the true intent of the JD, bypassing superficial keyword reliance.
  Stage 3 applies a mathematical Behavioral Multiplier, penalizing candidates with low recruiter response rates or high platform inactivity based on Redrob signals.
  Runtime is well under the 5-minute CPU limit, packaged in a glassmorphic Streamlit Command Center UI.

# ============================================================================
# Declarations
# ============================================================================
declarations:
  read_submission_spec: true
  code_is_original_work: true
  no_collusion: true
  honeypot_check_done: true
  reproduction_tested: true