This document outlines the strategic roadmap and upcoming technical features for InterviewAI.
Current Version: v2.3.0 | Last Updated: February 2026
Full design spec: CODING_ROUND_DESIGN.md
- Day 1 — Backend Brain: Build Problem Spec Generator, AI Dry Run Engine, and
/interview/coding-submit+/interview/coding-logendpoints. - Day 2 — Hint Engine + Personas: Tiered Hint Generator, Adinath's Pressure Mode, Veda's Verbalization Gate, and Resume Hook Extractor.
- Day 3 — Frontend Whiteboard UI: Split-pane editor + Persona toggle + Post-Round Code Review Report render.
- Add a "Researcher at Work" animated terminal in the UI to show real-time agent activity.
- Show the user when the AI is in "Stealth Mode" vs "Public Research" mode with a visual indicator.
- Polish the
discoveries.jsonvisual rendering in the dashboard.
- Implement a final "Hiring Decision" report after all 5 rounds are completed.
- Implement iterative difficulty scaling (Round 2 is harder if Round 1 was "Elite").
- Cross-Continental Localization: Enhance the Router to detect if a company has different processes (e.g., Google US vs Google India).
- Discovery Dashboard: Create a UI screen to browse the global
discoveries.jsondatabase. - Real-time Search Streaming: Push live research logs (ACCEPTED/REJECTED links) directly to the user's dashboard.
- Automatic Memory Pruning: Build a script to periodically "refresh" discoveries that are older than 6 months.
- Refine the "Critic" agent prompts to catch even more subtle hallucinations.
- Expand the Curated Database from 398 companies to 500+.
- Implement robust error handling and retry logic for DuckDuckGo rate limits (exponential backoff).
- Build the complete logic to transform interview feedback into a personalized 7-day study plan stored in the database, currently only the generator exists.
- Implement real-time WebRTC logic to analyze body language and tone during the session.
- Create a global leaderboard for "Google-Level" readiness scores.
- Allow users to share their "Interview Intelligence Reports" on LinkedIn.
- Mermaid Architecture Diagram — Fixed parse error on GitHub. All node labels properly quoted.
- Docs Reorganization — All loose markdown files moved to
/docsfolder with navigation index. - Evergreen Perpetual Freshness (v2.2.2) — Dynamic year calculation via
datetime. No more hardcoded years. - Geographic Guardrails — Location-aware
router_nodeprevents cross-continental name collisions. - Domain Guard (v2.2) — Prevents AI from forcing LeetCode rounds on non-tech companies.
- API Key Security — Rotated exposed keys, cleaned Git history, secured in
.env. - Neon PostgreSQL — Cloud database connected and used for session + user storage.
- Generic Article Purge (v2.3) — Python domain blocklist strips datacamp/guru99/intellipaat junk from Researcher results before Auditor sees them. Fixed Grant Thornton Bharat save failure.
- Domain-Aware Round Engine (v2.3) — Complete rewrite of
round_config.py. All 12 domains now get correct domain-specific rounds. Finance → Case Study. Healthcare → Situational. Creative → Portfolio Review. Tech → System Design. - Managerial Round Correction (v2.3) — Managerial round now applies to ALL domains at Senior level. Was incorrectly restricted to Tech + Business only.
- Domain Category Priority Fix (v2.3) — Fixed
generate_domain_report.pymapper order bug whereAccounting/Consultingwas binned under Business instead of Finance.