Assess evidence, not vocabulary
The experience should look for relevant responsibilities, outcomes and transferable patterns rather than rewarding surface-level keyword overlap.
AI-Enabled Decision Support
Rohit’s AI Navigator
Turning role-fit ambiguity into a clearer, evidence-led conversation.
A live AI-enabled role-fit assessment designed to help hiring teams examine alignment, relevant evidence, potential gaps and areas worth exploring in conversation — without pretending the system should make the hiring decision.
Live ProductEvolving Case Study
- Product type
- AI-enabled role-fit assessment
- Primary user
- Recruiters and hiring teams
- Role
- Product concept, judgment model, experience design and validation direction
- Current state
- Live and evolving
Overview
The problem was not generating more candidate commentary.
The harder problem was helping a hiring team examine fit with enough structure to be useful, enough evidence to be credible and enough humility to keep the final judgment human.
- Context
- Job descriptions combine strategic expectations, domain requirements, delivery responsibilities and organizational signals that are difficult to assess through keyword matching alone.
- Opportunity
- Create a decision-support experience that translates complex role requirements into a clearer assessment of alignment, evidence, gaps and useful follow-up questions.
- Product principle
- Support a better conversation rather than produce an automated hiring verdict.
The Decision Problem
Fit is contextual, but the first review is often shallow.
- Job descriptions vary widely in clarity and quality.
- Candidate experience is distributed across roles, industries and outcomes.
- Keyword matching can miss transferable evidence.
- Confident narrative can hide weak evidence.
- Hiring teams need both strengths and gaps, not only a positive summary.
- A useful assessment should help generate better interview questions.
The product needed to create structure without creating false certainty.
Where The Judgment Showed Up
Four decisions shaped the product.
Show strengths and gaps together
A credible assessment should help the user understand both the strongest alignment and the areas that require validation or deeper conversation.
Make the output actionable
The result should provide useful areas to explore, not simply a score or generic summary.
Keep the final call outside the system
The product supports interpretation. It does not possess the organizational context, accountability or human understanding required to make the hiring decision.
Experience Design
A complex assessment needed a simple path.
- Stage 01
Provide the role
The user supplies the job description or role context they want to examine.
- Stage 02
Interpret the expectations
The product organizes the role into meaningful expectations rather than treating it as an undifferentiated block of text.
- Stage 03
Connect relevant evidence
The experience examines where Rohit’s background and outcomes provide credible alignment and where evidence is weaker or incomplete.
- Stage 04
Support the conversation
The result surfaces strengths, gaps and areas worth testing through recruiter or hiring-manager discussion.
Technical Architecture
Architecture
React and TypeScript front end connected to Supabase Edge Functions and a structured candidate-evidence layer. Versioned evidence, retrieval and verdict guardrails separate AI interpretation from deterministic decision logic, while public-response controls keep private provenance and internal metadata out of the experience.
Human Boundaries
The product supports judgment. It does not own it.
No automated hiring decision
The user remains responsible for determining whether the candidate should progress.
No claim of complete context
The assessment is constrained by the role description and available candidate evidence.
No confidence without challenge
Users should investigate unsupported certainty, missing evidence and role-specific assumptions.
No substitute for conversation
The most valuable next step is often a better recruiter or hiring-manager discussion.
An AI-assisted assessment can improve preparation. It cannot understand every organizational reality surrounding the role.
Validation Approach
Evaluate usefulness before claiming intelligence.
- Review assessments across different role types and seniority levels.
- Check whether evidence is attached to the correct experience.
- Check whether meaningful gaps remain visible.
- Identify unsupported certainty or overly generous interpretation.
- Compare whether the output helps generate better discussion questions.
- Refine product language and interaction based on observed confusion or misuse.
Evaluation and validation
The assessment was reviewed against eight frozen job-description fixtures spanning aligned, adjacent and hard-negative roles. Reference outcomes covered all three verdict levels: Strong Fit, Worth a Conversation and Probably Not Your Person.
Automated regression checks verified benchmark behaviour and repeatability, while expert review identified where explanations were too generic, evidence attribution needed strengthening or confidence exceeded the available evidence.
What I Learned
Better AI products make their limits visible.
- 01
The quality of the input shapes the credibility of the assessment.
- 02
A balanced result is more useful than an enthusiastic result.
- 03
Users need reasons and discussion prompts, not only conclusions.
- 04
Product trust depends as much on boundaries and language as on the underlying system.
What Evolves Next
The next work is deeper evidence, evaluation and product clarity.
Evaluation
Develop a more explicit review approach for evidence quality, unsupported claims, gaps and usefulness.
Experience
Improve how role expectations, supporting evidence and discussion areas are presented.
Product boundaries
Continue clarifying what the product can support, where uncertainty remains and when human review must override the output.
Why This Product Matters
AI becomes useful when it improves the decision around it.
Rohit’s AI Navigator demonstrates how AI can support a practical, ambiguous decision while keeping evidence, uncertainty and human accountability visible.