Recto/submission_metadata.yaml at main · Aryanbhargava18/Recto · 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
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
# Redrob Hackathon — Submission Metadata
# This file mirrors portal metadata as required by submission_spec Section 10.3.

# ============================================================================
# Team identity
# ============================================================================
team_name: "Recto"

primary_contact:
  name: "Aryan Bhargava"
  email: "aryan.2024@nst.rishihood.edu.in"
  phone: "7888609660"

team_members:
  - name: "Aryan Bhargava"
    email: "aryan.2024@nst.rishihood.edu.in"
    role: "Ai student"

  - name: "Harshit Kudhial"
    email: "harshitkudhial@gmail.com"
    role: "Ai student"

# ============================================================================
# Code and reproducibility
# ============================================================================
github_repo: "https://github.com/harshitnub077/Recto"

sandbox_link: "https://rectoo.streamlit.app"
# Acceptable: HuggingFace Spaces, Streamlit Cloud, Replit, Colab, Docker

reproduce_command: "python main.py --candidates ./candidates.jsonl --output ./results"
# Runs end-to-end within 3 minutes on CPU, 16GB RAM, no network.

# ============================================================================
# Compute environment
# ============================================================================
compute:
  platform: "MacBook Air M2"
  cpu_cores: 8
  ram_gb: 16
  python_version: "3.13"
  os: "macOS 15 Sequoia"
  uses_gpu_for_inference: false
  has_network_during_ranking: false
  pre_computation_required: true
  pre_computation_time_minutes: 3
  pre_computation_description: |
    TF-IDF sparse similarity matrix pre-computed via precompute_embeddings.py.
    This generates a semantic_scores.pkl file used as an additive scoring boost.

# ============================================================================
# AI tools declaration
# ============================================================================
ai_tools_used:
  - "Claude"
  - "Cursor"
  - "Gemini / Antigravity"

ai_usage_summary: |
  Used Claude and Gemini for architecture discussion, code review, and debugging.
  Used Cursor for IDE autocomplete and refactoring.
  No candidate data was sent to any LLM during ranking.
  The ranking pipeline execution is fully deterministic with zero external API calls.

# ============================================================================
# Approach summary
# ============================================================================
methodology_summary: |
  Deterministic, rule-based ranker with pre-computed TF-IDF semantic boost.
  Five-layer scoring pipeline:
  (1) Hard-kill filters eliminate salary traps and honeypot keyword stuffers.
  (2) Core IR scoring parses career history for actual retrieval roles,
      applies rare-skill matching (BM25, Learning to Rank, ColBERT, etc.),
      and uses a coherence ratio to penalize keyword stuffers.
  (3) Behavioral multipliers weight response rates, notice periods,
      recruiter saves, and ghost detection.
  (4) Pre-computed TF-IDF cosine similarity provides an additive semantic
      boost without overriding strong heuristic candidates.
  (5) Deterministic sort by unified HYBRID_SCORE ensures zero inversions.
  Total runtime: ~3 minutes for 100K candidates on CPU. Peak RAM: ~4GB.

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