Start with the decision
Define who is making the decision, what evidence they need and what uncertainty the product should reduce before deciding whether AI belongs in the experience.
AI in Practice
AI as decision support, with judgment kept human.
I use AI to accelerate research, synthesis, prototyping and structured decision support — but not to outsource accountability, context or the final call.The product challenge is rarely whether AI can generate an answer. It is whether the experience helps a person make a better, more responsible decision.
Product Principles
The questions come before the technology.
Use AI where interpretation adds value
AI is most useful when the work involves synthesis, comparison, pattern recognition or explanation — not when a deterministic interface would be clearer and safer.
Make the reasoning inspectable
A useful decision-support product should help the user understand why an assessment was produced, what evidence supports it and where uncertainty remains.
Keep accountability human
The system may organize evidence and surface considerations. The person remains responsible for interpreting context, challenging the output and making the decision.
Where AI Helps
Acceleration without pretending uncertainty disappeared.
- Synthesizing complex inputs
- Comparing role expectations with relevant evidence
- Surfacing patterns, omissions and areas for discussion
- Producing structured first drafts for human review
- Accelerating research and product discovery
- Exploring alternative framings before committing to a direction
- Supporting prototype validation and iteration
Where Judgment Matters
Context, consequence and accountability cannot be delegated.
- 01Interpreting incomplete evidence
- 02Understanding organizational context
- 03Recognizing when a recommendation is too confident
- 04Weighing fairness, risk and downstream consequences
- 05Making the final hiring, investment or product decision
AI can improve the quality and speed of preparation. It does not remove responsibility from the person acting on the result.
Built Proof
Rohit’s AI Navigator
AI-enabled role-fit assessment
A live recruiter-facing decision-support experience designed to help hiring teams examine role alignment, relevant evidence, potential gaps and areas worth exploring in conversation.
- Decision problem
- Hiring teams often need to interpret broad job descriptions against complex candidate experience, while avoiding both keyword matching and unsupported certainty.
- Product response
- Create a structured assessment experience that organizes alignment, evidence, gaps and useful follow-up questions into a clearer decision surface.
- Human boundary
- The product does not make the hiring decision. It supports a more informed conversation and gives the user reasons to challenge, investigate or validate.
- Current state
- Live and evolving, with the product story, validation approach and supporting evidence continuing to develop through real use and structured review.
Responsible Product Questions
What has to be true before an AI feature earns trust?
- 01
Is AI improving the decision or merely adding novelty?
- 02
Can the user understand the basis of the output?
- 03
What happens when the available evidence is incomplete?
- 04
Where could confidence exceed what the evidence supports?
- 05
Who remains accountable when the recommendation is wrong?
Responsible AI product work is not a one-time disclaimer. It is an ongoing design, evaluation and operating responsibility.
How The Practice Evolves
Build, observe, challenge, refine.
- Step 01
Frame
Clarify the decision, user, evidence and risk.
- Step 02
Prototype
Build the smallest experience capable of testing the product assumption.
- Step 03
Evaluate
Review output usefulness, failure modes, trust and decision quality.
- Step 04
Refine
Improve the product, guardrails and operating approach based on evidence.
AI With Purpose
Need AI to support a real product decision?
I help teams identify where AI can create useful leverage, where simpler systems are better and where human accountability must remain explicit.