CoverCraft AI by Ambuj Kumar Tripathi

---

🎯 The Engineering Philosophy

❌ The "Toy AI Generator" Fallacy

99% of cover letter tools on the internet are simple prompt wrappers that cause immediate disqualification during technical recruiter screens:

• Hallucinated Metrics: They invent fake quantitative claims (*"Increased revenue by 43% using Docker"* when the candidate only listed Docker under skills).
• Fluff & Sycophancy: They inject generic, embarrassing praise (*"I have long admired your revolutionary, industry-defining synergy"*).
• Zero Company Grounding: They have no real-time awareness of recent engineering pivots, active tech stacks, or corporate subsidiaries.
• Hidden Third-Party Traps: They fail to detect when a job posting is a third-party staffing/payroll agency disguised as a direct employer.

✅ The CoverCraft Architecture: Verifiable Evidence Graph

CoverCraft is not a template filler. It is an Evidence-Grounded Career Workspace & Claim-Validation Engine designed for senior engineers and discerning hiring teams.

Candidate Resume (PDF)          Job Description (JD)           Live Web (Tavily + Jina AI)

│ │ │ ▼ ▼ ▼ ┌──────────────────┐ ┌──────────────────┐ ┌────────────────────┐ │ Deterministic │ │ Deterministic │ │ 8-Page Multi-Board │ │ Text & Evidence │ │ Fallback Regex │ │ Crawl & Markdown │ │ Extraction Layer │ │ & Vendor Auditor │ │ Deep Extraction │ └─────────┬────────┘ └────────┬─────────┘ └─────────┬──────────┘ │ │ │ └─────────────────────────────┼─────────────────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Human-in-the-Loop Gateway │ │ (Source Approval & Selection) │ └───────────────┬─────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Evidence-Grounded Generator │ │ • Strict Verbatim Anchors │ │ • Evidence-Grounded Claim Validation │ │ • Citation Footnotes [1], [2] │ └───────────────┬─────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Interview-Defensible Output │ │ (Every claim backed by facts) │ └─────────────────────────────────┘

---

⚡ Core Architectural Pillars

1. Multi-Source Real-Time Intelligence & Deep Markdown Crawling

• Tavily 8-Page Authoritative Crawl: Queries live enterprise footprints targeting careers, linkedin, naukri, and ambitionbox.
• Jina AI Reader (`r.jina.ai`) Deep Scraping: Strips HTML clutter, popups, and cookie walls, returning clean, high-fidelity markdown with an exponential timeout fallback.
• Three-Tier Domain Classifier: Categorizes incoming intelligence into Company Official, Job Portals & Reviews, and External Engineering News.

2. Deterministic Fallback Regex & 4-Tier Company Intelligence

When LLM JSON structures are delayed or partially populated, a deterministic regex pipeline extracts ground truth directly from raw JD tokens:

• 📋 Job Context Sub-Tab (Default): Extracts Role, Work Model (Remote / Hybrid / On-Site), Experience Band, and Disclosed Salary Status.
• ⚠️ Third-Party / Staffing Detection: Flags staffing vendors, C2H (Contract-to-Hire), and payroll intermediaries with verbatim quotes directly extracted from the JD.
• 🏢 Company Identity Sub-Tab: Verifies whether the firm is a Direct Employer, Services Vendor, GCC (Global Capability Center), or Product Company. Missing fields strictly report "Not verified from available sources".
• ⚡ Hiring Signals Sub-Tab: Isolates technology focuses, active initiatives, and recent hiring surges with cited references [1], [2].
• 🔗 Sources Coverage Sub-Tab: Displays full URL citations and scrape statuses.

3. Evidence-Grounded Generation, UX & Claim Validation Policy

• Evidence-Grounded Synthesizer: System instructions enforce that every single achievement claim in the cover letter must map directly to a verified bullet in the candidate's resume.
• Interactive In-Line Evidence Markers with Clean Submission Export: On-screen interactive badges ([Resume Anchor] & [Source Citation]) allow candidates to hover and verify exact source quotes with a live [Evidence Markers: ON/OFF] toggle. When candidates click Copy to Clipboard or Download Markdown/PDF, internal citation tags are automatically stripped to deliver 100% clean, submission-ready letters for recruiters.
• Adversarial Overclaim Red-Teamer (Veracity Audit Gate): Scans generated text against candidate resume verbs to catch senior inflation (e.g., resume states "assisted team with Kubernetes rollout" vs draft stating "spearheaded enterprise migration"). The engine automatically softens unverified verbs to authentic baseline evidence and records an audit log.
• Live Step Progress Tracker & Dynamic Elapsed Timer: An event-driven 4-phase execution tracker (01. Competency Match → 02. MCP Company Research → 03. Human Approval → 04. Evidence Synthesis) with a live counting timer (00:04s) and active tool indicator so candidates clearly follow live execution without confusion or perceived stalls.
• Human-in-the-Loop (HITL) Gate: Users explicitly inspect, evaluate credibility scores, and toggle on/off scraped web sources before any external intelligence enters the LLM generation prompt.

4. Native Model Context Protocol (MCP) Integration (8 Production Tools)

[!TIP]

🛡️ Triple-Registry Verified FastMCP Server (Glama · Smithery · M8ven):

CoverCraft's FastMCP server and its 8 registered tools have been independently audited and verified across leading registries with Glama Grade A - A, M8ven Trust Score 89/100 (Grade B), and Smithery Verified (8 Tools) with continuous live monitoring and 0 security vulnerabilities.

• Contains a standalone Anthropic Model Context Protocol (MCP) server written in Python 3.