SignalHire/submission_metadata.yaml at main · DevanshSrajput/SignalHire · 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
# Redrob Hackathon — Submission Metadata

# ============================================================================
# Team identity
# ============================================================================
team_name: "SignalHire"

primary_contact:
  name: "Devansh Rajput"
  email: "rachitmittalxc@gmail.com"
  phone: "+91-6203454887"

team_members:
  - name: "Devansh Rajput"
    email: "rachitmittalxc@gmail.com"
    role: "Team Lead"

# ============================================================================
# Code and reproducibility
# ============================================================================
github_repo: "https://github.com/DevanshSrajput/SignalHire"

sandbox_link: "https://huggingface.co/spaces/DevanshSrajput/SignalHire"

reproduce_command: "python rank.py"
# rank.py reads candidates from data/candidates.jsonl (mount or copy there)
# and writes output/submission.csv. Precomputed artifacts are baked into
# the Docker image — no network or GPU needed at ranking time.

# ============================================================================
# Compute environment
# ============================================================================
compute:
  platform: "Linux x86_64"
  cpu_cores: 8
  ram_gb: 16
  python_version: "3.10"
  os: "Ubuntu 22.04 LTS"
  uses_gpu_for_inference: false
  has_network_during_ranking: false
  pre_computation_required: true
  pre_computation_time_minutes: 30   # GPU: ~4 min; CPU: ~30 min

# ============================================================================
# AI tools declaration
# ============================================================================
ai_tools_used:
  - "Claude"

ai_usage_summary: |
  Used Claude for architecture discussion, code review, and debugging.
  No candidate data was fed to any LLM. All scoring logic, weights, and
  disqualification rules are original work.

# ============================================================================
# Approach summary
# ============================================================================
methodology_summary: |
  SignalHire ranks 100K candidates via a five-signal composite score:
  Technical Fit (0.35), Career Quality (0.25), Availability Signal (0.20),
  Seniority Fit (0.12), and Semantic Similarity (0.08). Embeddings are
  precomputed offline with all-MiniLM-L6-v2; ranking is a single matrix
  multiply at inference time (<1s). A disqualification layer hard-removes
  honeypots (claimed YoE exceeds career timeline), ghost profiles, and
  pure-research candidates before scoring. Soft penalties apply for
  all-consulting careers, no-code-18mo, and CV/robotics-only profiles.
  Evidence-cited per-candidate reasoning cites the exact skill or sentence
  that triggered each score component.

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