Honestify
← Back
AI Improvementsv1.8.0

Better AI Answer Quality

Better AI Answer Quality (2026-03-12): Answer quality improvements: stricter grounding, metric extraction from projects, and reduced hedging in interview-style responses.

·4 min read·Honestify Engineering

Feature area: Answer Quality

On this page

This release covers Better AI Answer Quality Version 1.8.0, shipped 2026-03-12. Status: shipped. No breaking changes.

Summary

Answer quality improvements: stricter grounding, metric extraction from projects, and reduced hedging in interview-style responses.

Users reported AI answers were 'safe but vague'—heavy on frameworks, light on their specific trade-offs and numbers. Retrieval reranking, mandatory project citation when available, and post-processing to surface quantified bullets from profile data.

What Changed

Reranking

Improved

Cross-encoder reranker on retrieved KB chunks.

Metric surfacing

New

Extract latencies, team sizes, and scale figures from structured fields.

Anti-hedge rules

Improved

Prompt constraints reduce 'it depends' without trade-off articulation.

Interview-ready rating

Before

54%

After

76%

Post-session survey (n=310)

  • Reranking — Cross-encoder reranker on retrieved KB chunks.
  • Metric surfacing — Extract latencies, team sizes, and scale figures from structured fields.
  • Anti-hedge rules — Prompt constraints reduce 'it depends' without trade-off articulation.

Why We Built It

Users reported AI answers were 'safe but vague'—heavy on frameworks, light on their specific trade-offs and numbers.

We prioritized this work because user interviews consistently surfaced this as a blocker to daily use. The fix needed to be durable—not a patch—so we addressed root causes in Answer Quality rather than symptoms alone.

Engineers, recruiters, and hiring managers all benefit when Honestify behaves predictably in production. This release reflects that bar.

User Impact

User rating 'Would use in real interview' increased from 54% to 76%.

AudienceHow you benefit
EngineersFaster profile setup, clearer AI answers, less manual rework
RecruitersMore complete profiles and reliable share links when candidates use Honestify
Founders / hiring managersBetter signal on candidate preparation and skills alignment
Platform engineersPerformance headroom for upcoming features

Relevant skills: rag, prompt engineering, embeddings, python. Target roles: ai engineer, backend engineer, full stack engineer.

Technical Highlights

  • Reranker model hosted on inference endpoint
  • Structured field priority in context assembly
  • Answer post-processor for citation formatting
  • Quality sampling to human review queue

Prompt and retrieval changes versioned separately with eval gating before production promotion.

Before

Better AI Answer Quality: before vs after

Before

Answers often opened with generic definitions before touching user-specific detail.

After

Lead with user's project context; definitions only when explicitly requested.

Users moving from the previous experience should notice lead with user's project context; definitions only when explicitly requested.

Screenshots

Future Improvements

What we are building next

  • Answer diff vs user edit for learning
  • Per-question quality scores
  • Multi-model fallback on timeout

Known limitations

  • · Quality gains require populated project metrics fields

Feedback welcome: Reply via in-app feedback or support—especially if you hit edge cases we did not cover in this release.

This update connects to other Honestify work:

Create your own AI profile

Upload your resume, add expertise, and share a profile link beside LinkedIn so recruiters can ask follow-up questions before the interview.

Create your AI profileUpload your resume