rightfit.so

Hiring managers at AI-first startups waste 1,800 hours/year manually reviewing candidates—reading resumes, intuiting fit, and drafting interview questions without structured frameworks (source: rightfit.so time-tracking study, n=112 managers, Q1 2025). At 50 candidates/role and 8 roles/year, this consumes 15 minutes/candidate—delaying hires and causing 32% inconsistent evaluations (source: internal audit of 500 rejected candidates).

The business case:
2,000 active hiring managers × 400 candidate reviews/year × $0.83/minute (blended $50/hr rate) × 13 minutes saved = $8.66M/year recoverable time
(Source: Manager count from CRM, review volume from 2024 user survey, wage from AngelList startup comp benchmarks)
If adoption is 40% of estimate: $3.46M/year
This excludes $220K/hire cost of bad-fit candidates avoided through structured red-flag detection.

This is an AI-generated candidate brief delivering skill match scores, culture signals, red flags, and gap-specific interview questions in <10 seconds. It is not an automated hiring decision-maker—human judgment remains central.

Greenhouse solves this through manual scorecard templates; Ashby via pipeline analytics dashboards; Seekout with keyword-based resume search.

WHO / JTBD: When a startup engineering lead reviews an ML candidate post-sourcing, they need to rapidly assess technical fit and cultural alignment against their AI product’s niche requirements—without manually cross-referencing resumes, GitHub, and behavioral cues for 15 minutes per candidate.

CURRENT BREAKAGE:

• End User: Scattershot review: 73% miss key skill gaps (e.g., "Deployed PyTorch models at scale") buried in resumes (source: user testing, n=28)
• Buyer (Founder): Inconsistent evaluations cause 22-day role-fill delays (source: Q4 customer churn survey) and 28% mis-hire rate for AI roles
• Business: $18K sunk cost per mis-hired candidate (source: Rightfit ROI calculator)
• Market: Losing deals to ATS platforms promising "AI screening" without vertical-specific logic

┌───────────────────────────────────────────────────────────────┐
│ PASTE PROFILE                                                 │
├───────────────────────────────────────────────────────────────┤
│ [Job Description Text Area] ▾  [Paste from LinkedIn]          │
│ [Candidate Profile Text Area] ▾ [Paste resume/CV]             │
│                                                               │
│                       [Generate Fit Brief]                    │
└───────────────────────────────────────────────────────────────┘

┌───────────────────────────────────────────────────────────────┐
│ AI FIT BRIEF: Elena Rodriguez - ML Engineer                   │
├───────────────────────────────────────────────────────────────┤
│ SKILL MATCH: 88%                                              │
│   ✅ STRONG: PyTorch, model deployment, TFX                   │
│   ⚠ GAP: Vertex AI, RL production experience                  │
│                                                               │
│ CULTURE SIGNALS                                               │
│   🔴 Ownership: High (3 shipped products)                     │
│   🟢 Collaboration: Medium (1 co-authored paper)              │
│                                                               │
│ RED FLAGS                                                     │
│   ‼ 11-month avg tenure (vs. 26mo industry)                   │
│                                                               │
│ INTERVIEW QUESTIONS                                           │
│   "Walk me through your most complex RL deployment"           │
│   "How do you debug model drift without ground truth?"        │
│                                                               │
│ RECOMMENDATION: Proceed with deep-dive                        │
└───────────────────────────────────────────────────────────────┘

Core Flow: Paste → Analyze → Brief → Decide. Uses ensemble model: ResumeBERT for skill extraction, fine-tuned GPT-4 for narrative analysis, custom classifier for culture signals. Rejected "auto-reject" functionality to preserve human agency. Output editable pre-interview.

Phase 1 — MVP (5 weeks)
US#1 — Brief Generation

• Given JD and candidate profile pasted
• When user clicks "Generate Fit Brief"
• Then display structured brief within 8 seconds with:
  • P0: Skill match % with 100% consistency (zero failures)
  • P1: ≥3 culture signals at ≥99.5% accuracy
  • P1: ≥1 red flag if pattern detected (e.g., job-hopping)
  • P1: ≥2 interview questions targeting largest skill gap
• If timeout >15s: Show "Try again" with offline fallback
• Validated by QA against 200 real AI-resume pairs

Out of Scope (Phase 1)

Phase 1.1 (3 weeks): Brief sharing + comment threading
Phase 1.2 (4 weeks): LinkedIn profile auto-import

Not Measuring:

• Briefs generated (vanity; doesn’t indicate usage depth)
• Raw "match score" (easily gamed; focus on gap reduction)
• NPS (lagging; time savings is leading indicator)

Risk: Hallucinated red flags (e.g., falsely claims candidate lacks PyTorch)
Probability: Medium | Impact: High
Mitigation: Rule-based validator for P0 attributes (skills/titles). Owner: ML Lead by W3.
────────────────────────────────────────────────
Risk: GDPR violation for EU candidate data
Probability: High | Impact: Critical
Mitigation: Anonymize pre-processing; legal sign-off on flow. Owner: CPO by W2.
────────────────────────────────────────────────
Risk: Competitor clones feature in 60 days (e.g., Ashby)
Probability: Medium | Impact: Medium
Mitigation: Patent pending on culture signal extraction. Owner: Counsel by W6.
────────────────────────────────────────────────
Risk: Managers over-rely on AI recs
Probability: Low | Impact: High
Mitigation: Mandatory "override reason" for non-recommended hires.

Kill Criteria:

1. False negative rate >5% at D30
2. Review time reduction <35% at D60
3. <50% of briefs lead to interview notes

Pilot (2 weeks):

• 15 high-volume customers (50+ hires/yr)
• Measure time savings and false flag rate
  GA Rollout:
• Tier 1 (100 customers): Opt-in with onboarding checklist
• Tier 2 (all): Feature flag enabled after 7 days
  Critical Path:
• Legal sign-off (W4)
• Inference load testing (W5)

Decision: Depth vs. breadth of skill taxonomy
Choice: 120 AI-specific competencies (e.g., "Transformer fine-tuning") over generic 500+ skill library
Rationale: Startup AI roles require niche expertise; generic terms ("Python") create false positives.
────────────────────────────────────────────────
Decision: Handling unverifiable claims (e.g., "built GPT-4 competitor")
Choice: Flag with ‼ without penalizing score
Rationale: Avoid over-indexing on resume puffery; surface for interview probing.
────────────────────────────────────────────────
Decision: Culture signal framework
Choice: Startup-specific traits (Ownership, Velocity, Grit) over corporate values
Rationale: "Innovation" scores are meaningless; shipping velocity correlates with startup success.
────────────────────────────────────────────────
Decision: Data retention policy
Choice: Briefs persist 30 days; raw resumes deleted post-processing
Rationale: Balances reusability with GDPR candidate privacy requirements.

Before: Maya (Eng Lead, Series A AI infra) pastes an ML candidate’s resume into her notes. She scans for keywords, misses the lack of cloud deployment experience, and drafts generic questions. After 18 minutes, she’s unsure—schedules a screening "just in case."

After: Maya pastes the JD and resume. In 6 seconds, she sees the 34% infra gap and question: "Describe an ML pipeline you containerized." She probes this in the interview, uncovering dealbreaker gaps. Total time: 3 minutes.

Pre-Mortem:
"It’s 6 months post-launch and adoption flatlined. Why?

1. False flags eroded trust—users reverted to manual reviews after 2 errors
2. We prioritized enterprise integrations over core accuracy, missing our wedge
3. Ashby shipped a 'Gap Finder' using our UX pattern 45 days post-launch"

Success looks like: Hiring managers start calls with "I saw your Vertex AI gap—let’s dig in." Founders report 30% faster technical interviews. Support tickets for "how to evaluate candidates" drop by half.