diff --git a/README.md b/README.md index acffebc..d1a5e76 100644 --- a/README.md +++ b/README.md @@ -4,7 +4,7 @@ --- -> 📚 **Project Docs:**  [Challenges](./docs/CHALLENGES.md)  |  [Next Steps](./docs/NEXT_STEPS.md)  |  [Coding Round Design](./docs/CODING_ROUND_DESIGN.md)  |  [Technical Flow](./docs/TECHNICAL_FLOW.md)  |  [Collaborators](./docs/COLLABORATORS.md)  |  [All Docs →](./docs/) +> 📚 **Project Docs:**  [Challenges](./docs/CHALLENGES.md)  |  [Next Steps](./docs/NEXT_STEPS.md)  |  [Coding Round Design](./docs/CODING_ROUND_DESIGN.md)  |  [Technical Flow](./docs/TECHNICAL_FLOW.md)  |  [All Docs →](./docs/) ---
diff --git a/docs/COLLABORATORS.md b/docs/COLLABORATORS.md deleted file mode 100644 index d85c76c..0000000 --- a/docs/COLLABORATORS.md +++ /dev/null @@ -1,91 +0,0 @@ -# 👥 Project Collaborators - -This document tracks the core team and the implementation history of the **InterviewAI** project. - ---- - -## 🏗️ Core Development Team - -| Name | Role | Focus | -| :--- | :--- | :--- | -| **Karan Shelar** | Lead Architect | AI Systems, LangGraph orchestration, Full-Stack Development | -| **Abhiraj** | Collaborator | Full-Stack Development (Backend & Frontend), System Refinement | - ---- - -## 📝 Activity & Implementation Log - -This section is for tracking specific task completions and problem-solving history. - -### **Karan** -- **Implemented** Agentic Discovery System using LangGraph (Researcher-Architect-Critic). -- **Implemented** Stealth Mode logic with JD reverse-engineering for unknown companies. -- **Implemented** Neon Cloud PostgreSQL migration and production database sync. -- **Implemented** Massive company intelligence database (383+ profiles). -- **Implemented** Universal Gemini evaluation logic with company-context integration. -- **Implemented** Zero-Hallucination Framework: Integrated Critic Grounding, Trust Score system (0-100), and "I Don't Know" Policy to prevent AI imagination. -- **Implemented** Optimization Suite: Added `rapidfuzz` for better matching, refined DDGS queries for noise reduction, and a "Status Stream" for latency feedback. -- **Validated** System performance against real-world firms (e.g., Idolize Business Solutions, Mistral AI). -- **Implemented** Source Attribution: The AI now captures and links all research sources in the final profile. -- **Implemented** Evergreen Temporal Search: Upgraded the Researcher to dynamically calculate years for perpetual freshness. -- **Implemented** The Auditor Agent (The Bouncer): A dedicated node that filters 'Vomit' (noise like Vastu/NCERT), verifies Company Identity vs Job Description, and generates an 'Audit Log' for developer transparency. -- **Fixed** Information Starvation Bug: Upgraded the Researcher/Auditor nodes to preserve and pass full search metadata (Title + Body). This slashed hallucinations and fixed the "Confidence 0" issue for rare companies. -- **Implemented** Domain Guard (Role-Company Logic): Added a contextual cross-check to the Auditor and Architect nodes. The system now detects mismatches between the user's role (e.g., Software Engineer) and the company's actual industry (e.g., Creative Agency), and provides a logical explanation of the alignment instead of hallucinating. -- **Improved** Validation Logic: Enforced strict Critic-based approval before any discovery is saved to `discoveries.json`, ensuring the memory stays pristine. -- **Updated** Comprehensive Documentation: Refactored root, backend, and discovery READMEs to reflect the new 4-agent architecture and Mermaid diagrams. Created `CHALLENGES.md` to document the project's complex technical solutions. -- **Fixed** Critical Data Integrity loophole: Implemented "Global Vault" isolation for Stealth/Synthetic data. -- **Fixed** Strict Identity Bug: Tightened fuzzy matcher threshold (95%) to prevent collisions. -- **Implemented** Dynamic Domain Intelligence (v2.2): Added logic to prevent 'Role Forcing' on non-tech companies (e.g., Clinical/Legal/Creative) during profile architecture. -- **Implemented** Geographic Guardrails: Built a location-aware `router_node` to prevent cross-continental naming collisions (e.g., MOC India vs Moffitt USA). -- **Implemented** Confidence Score System: Standardized confidence weighting (0-160) for clear Status/Reliability tracking. -- **Fixed** Memory Integrity: Recovered and cleaned `discoveries.json` after accidental malformed JSON writing; added a 98% fuzzy threshold for discovery memory retrieval. -- **Implemented** Audit Status: Added `ACCEPTED/REJECTED` status tracking in the Audit Log for every source link. - -- **Implemented** Evergreen Perpetual Freshness (v2.2.2): Replaced hardcoded years (2025/2026) with dynamic temporal logic that auto-calculates current and future years for research queries. -- **Fixed** Mermaid Diagram Rendering Error: Quoted all special-character labels in the README architecture diagram so it renders correctly on GitHub (was throwing 'Parse error on line 10'). -- **Refactored** Documentation Structure: Moved all loose root-level markdown files into a dedicated `/docs` folder using `git mv` (preserving history). Created a `docs/README.md` navigation index and a quick-link bar in the main README. -- **Designed** Coding Round Intelligence System (v3.0 Roadmap): Authored full 9-section blueprint (`CODING_ROUND_DESIGN.md`) covering Whiteboard Mode, AI Dry Run, Tiered Hints, JSON Learning Ledger, and Persona Architecture (Adinath vs. Veda). -- **Tested** Autonomous Discovery: Validated with companies including WABRIC, Aminuteman Technologies, and Shastra Solutions. Confirmed audit logs, source filtering, and confidence scores are working correctly. -- **Fixed (v2.3)** Generic SEO Article Contamination: Added Python-level domain purge list in Researcher Node (14 blocked domains). Fixed Grant Thornton Bharat, BDO India, and Nexia International India save failures caused by generic AI article pollution. -- **Built (v2.3)** Domain-Aware Round Engine: Completely rewrote `round_config.py` — all 12 career domains now get domain-specific interview rounds (Case Study for Finance, Situational for Healthcare, Portfolio Review for Creative, etc.). -- **Fixed (v2.3)** Managerial Round Scope: Corrected Managerial round to apply to ALL domains at Senior level (was wrongly restricted to only Tech + Business). -- **Fixed (v2.3)** Domain Report Priority Bug: Reordered `DOMAIN_MAPPER` in `generate_domain_report.py` so Finance is checked before Business, fixing `Accounting/Consulting` miscategorization. -- **Implemented (v2.4)** Multi-Round Transition Logic (The Marathon): Built the frontend overlay and backend flow to handle the jump between interview rounds (Option B). -- **Implemented (v2.4)** Executive Scorecard & Feedback Engine: Created a premium evaluation UI with STAR Method Analysis and Vibe Metering (Option C). -- **Implemented (v2.4)** Crowdsourced Self-Learning Memory: Built `MemoryService` and `stealth_registry.json`. The AI now learns from user sessions using "Witness vs Expert" consensus logic. -- **Optimized (v2.4)** Hybrid Duo Architecture: Standardized the local Llama-3 (Architect) + Gemini-2 (Communicator) duo for zero-latency, high-fidelity simulations. - -### **Abhiraj** -- *Starting contributions today (tracking development from this point forward).* - ---- - -## ✅ Major Version Milestones - -| Feature / Update | Version | Status | Key Highlights | -| :--- | :--- | :--- | :--- | -| **Executive Memory & Scorecard** | v2.4.0 | ✅ Done | STAR Analysis, Vibe Meters, Multi-Round Transitions, and Self-Learning Stealth Registry. | -| **Domain-Aware Round Engine** | v2.3.0 | ✅ Done | 12 domains get their own round sequences (Case Study, Situational, Portfolio Review, etc.). | -| **Managerial Round for All Domains** | v2.3.0 | ✅ Done | Managerial now applies at Senior level across all 12 domains, not just Tech + Business. | -| **Generic Article Purge** | v2.3.0 | ✅ Done | Python domain blocklist strips SEO junk before Auditor. Fixed Finance company save failures. | -| **Domain Report Priority Fix** | v2.3.0 | ✅ Done | Reordered DOMAIN_MAPPER so Finance/Accounting is checked before Business/Management. | -| **Coding Round Intelligence Design** | v3.0 (Planned) | 🚧 In Progress | Whiteboard Mode, AI Dry Run, Tiered Hints, Learning Ledger. | -| **Docs Reorganization** | v2.2.2 | ✅ Done | All docs moved to `/docs` folder. Nav bar added to README. | -| **Mermaid Diagram Fix** | v2.2.2 | ✅ Done | Fixed GitHub parse error by quoting all special node labels. | -| **Evergreen Perpetual Freshness** | v2.2.2 | ✅ Done | Dynamic `datetime` year anchoring — no more hardcoded years. | -| **Geographic Guardrails** | v2.2.1 | ✅ Done | Location-aware routing to stop name collisions. | -| **Dynamic Domain Guard** | v2.2.1 | ✅ Done | Prevents forcing tech rounds on non-tech firms. | -| **Public Deployment** | v3.0 (Planned) | 🔮 Not Started | Backend (Render/AWS Lambda) + Frontend (Vercel) — planned. | -| **Agentic Intelligence Brain** | v2.1.0 | ✅ Done | Integrated LangGraph Researcher-Architect team. | -| **Stealth Mode Logic** | v2.1.0 | ✅ Done | AI now reverse-engineers JD for private startups. | -| **Neon Cloud PostgreSQL** | v2.1.0 | ✅ Done | Cloud-hosted database connected for sessions and users. | -| **398 Company Profiles** | v2.1.0 | ✅ Done | Pre-loaded intelligence database across 12 domains. | -| **ATS Resume Analysis** | v2.0.0 | ✅ Done | Gemini-powered PDF parsing and matching. | -| **Persona System** | v1.5.0 | ✅ Done | Created Adinath & Veda simulation personalities. | - ---- - -## 🛠️ Contribution Guidelines -- **Branches**: Create a feature branch for every major update. -- **Commits**: Use descriptive messages (e.g., `feat: add stealth mode logic`). -- **Sync**: Always run `alembic upgrade head` after pulling changes. diff --git a/docs/README.md b/docs/README.md index 3dcf13c..acace9b 100644 --- a/docs/README.md +++ b/docs/README.md @@ -8,7 +8,6 @@ This folder contains all internal design documents, planning files, and engineer | :--- | :--- | | [`CHALLENGES.md`](./CHALLENGES.md) | Engineering challenges encountered and the agentic design patterns used to solve them. | | [`CODING_ROUND_DESIGN.md`](./CODING_ROUND_DESIGN.md) | Full design blueprint for the upcoming Coding Round Intelligence feature (Whiteboard Mode, AI Dry Run, Tiered Hints, Learning Ledger, Personas). | -| [`COLLABORATORS.md`](./COLLABORATORS.md) | Contribution history and milestone log for all collaborators on the project. | | [`NEXT_STEPS.md`](./NEXT_STEPS.md) | Strategic roadmap — short-term goals, long-term vision, and tech debt tracking. | | [`TECHNICAL_FLOW.md`](./TECHNICAL_FLOW.md) | End-to-end technical flow of the platform, from user input to AI simulation output. | | [`IMPLEMENTATION_PLAN.md`](./IMPLEMENTATION_PLAN.md) | Phase-wise implementation checklist for all major features. |