Evident — Code-Grounded Career Intelligence

Code-grounded career intelligence engine that replaces AI resume hallucinations with verifiable repository proof, commit provenance, and interactive defense simulation.

Comment

💡 Inspiration: The Trust Crisis in Tech Hiring

Every hiring season, millions of aspiring software engineers and students face a brutal paradox:

  1. The Generative AI Slop Dilemma: Candidates copy-paste job descriptions into ChatGPT, synthesizing resumes inundated with generic, unverifiable claims ("Architected distributed, fault-tolerant microservices with high scalability"). Recruiters, having developed automated cynicism, spend less than 6 seconds per resume and discount ungrounded text at scale.
  2. The Absence-of-Evidence Problem: Early-career builders, student hackers, and open-source contributors do not have an absence of work — they have an absence of verifiable evidence. Hundreds of intricate Git commits, university capstones, architectural trade-offs, and late-night pull requests sit scattered across GitHub branches. When applying, they either generate artificial AI buzzwords or severely undersell what they actually built.

"Your resume should describe what you can prove. Evident remembers what you have done. Apply knows when it matters."
Brand Thesis: Don't Claim It. Prove It.

We created Evident to eradicate AI resume hallucination by establishing an unbroken, verifiable chain of custody between raw Git commits and professional career claims.

⚡ What It Does

Evident is a Code-Grounded Career Intelligence Engine that continuously analyzes connected Git repositories, constructs a dynamic Interactive Evidence Graph, and grounds every professional claim in concrete commit history and source code syntax trees.

🌟 Core Capabilities & The 4-Stage Loop

  • 1. Living Evidence & Interactive Force Graph: Rather than static text lists, candidate capabilities are rendered as an interactive, relational force-directed canvas. Skills, projects, and commits exist as interconnected nodes with live physics, visual telemetry, and instant node inspection.
  • 2. Transparent Opportunity Matching: Ingests any real-world job posting and outputs an objective Evidence Coverage Matrix (such as 5 Direct, 2 Supported, 1 Gap) rather than an arbitrary, opaque match percentage. Critical skill gaps are surfaced honestly alongside remediation roadmaps.
  • 3. Application Studio & The 1-Tap Provenance Inspector (The Hero): Synthesizes grounded resume bullets linked directly to retrieved evidence. Tapping [Why this claim?] slides up an instant mobile provenance drawer revealing:
    • Target repository (RIFT, KALMAN, SYNTRA)
    • Relative source file path (src/auth/tokenRotation.ts)
    • Matching Git Commit SHA (b81c44a)
    • Syntax-highlighted code extract proving first-party authorship
  • 4. Architectural Defense Arena: Moving beyond static resumes, Evident prepares candidates for real technical interview grillings by generating questions derived directly from their own code ("Why did you choose database-backed token rotation in tokenRotation.ts instead of stateless JWTs?"), scoring their trade-off defense across technical rigor and failure-mode preparedness.
  • 5. Cryptographic ProofPack Export: Packages verified evidence and commit proofs into an immutable, portable dossier with SHA-256 integrity checksums for technical recruiters and hiring committees.

📐 The 5-Dimensional Evidence Model

Rather than collapsing candidate competency into a subjective number, Evident models technical evidence across five orthogonal dimensions:

Evidence Vector = { Status, Artifact, Authorship, Confidence, Freshness }

  • 1. Evidence Status: Direct Source (first-party algorithmic logic and API routes), Supported (build configs and manifests), Partial (adjacent conceptual paradigms), Evidence Gap (zero signal detected), and User-Declared.
  • 2. Source Artifact: Exact pointer to Git commit hash, syntax AST node, and relative file path.
  • 3. Authorship Attribution: Filters out cloned boilerplate; weights signals by author contribution diffs (1.0 for primary author, 0.6 for collaborator).
  • 4. Confidence Tier: Deterministic heuristic calculated from code line density, test coverage, and implementation depth.
  • 5. Temporal Freshness: Exponential decay modeling recency based on time elapsed since the last verified commit.

Objective Coverage Metric Formula

Coverage Score = [ (Weighted Signal Strength Sum) / (Total Requirement Weights) ] × Must-Have Penalty Multiplier

Where requirements carry weighted priorities (2.0 for must-have, 1.0 for preferred), evidence strength scales objectively from 1.0 (Direct Source) to 0.0 (Gap), and missing critical requirements automatically apply a 35% gap penalty to prevent artificial inflation.

🛠️ How We Built It

Evident is engineered to the highest production standards, adhering strictly to Clean Architecture (Robert C. Martin) and A Philosophy of Software Design (John Ousterhout):

System Architecture Pipeline

  • Ingestion & Normalization: The GitHub ingestion service indexes Git trees, extracts commit logs, and parses AST signals into normalized evidence fixtures.
  • Core Domain Layer (Pure TypeScript): Framework-agnostic domain modules including proofGraph (relational node linking), matchingEngine (deterministic scoring), and claimAuditor (anti-hallucination verification).
  • Deep RevenueCat Integration (react-native-purchases): Native commerce engine powered by PurchaseService and our subscription state store:
    • Living Career Memory Entitlement (evident_pro): Monetizes continuous background repository sync, unlimited Architectural Defense interview rounds, and cryptographic ProofPack exports.
    • Tier Packaging: Annual Career Pass ($49.99/year, Save 48% with trial) for continuous yearly monitoring, and Monthly Career Sprint ($7.99/month) for active recruitment windows.
    • Zero-Friction Judge Sandbox Controller: Seamlessly switches between live RevenueCat SDK configuration and a local Test Store state machine, allowing judges to test purchases and restores with 1 tap and zero credential friction.
  • Presentation & Experience: Built with React Native and Expo (SDK 54) using TypeScript. Features an autonomous Ambient Aurora shader background, Apple-style frosted glass cards, and Alan Cooper's rich floating undo toasts in place of jarring modal popups.

🚧 Challenges We Ran Into

  • The Anti-Hallucination Barrier: Generative AI models naturally tend to invent credentials to please the user. We engineered a strict, multi-pass Claim Auditor. Any synthesized bullet asserting skills absent from the proof graph (e.g., claiming Kubernetes or AWS without commit provenance) is rejected and flagged as an unverified gap.
  • Mobile Relational Graph Physics: Rendering interconnected multi-repo dependency graphs on mobile devices without frame drops required designing a lightweight, flattened layout engine with spatial memoization and decoupled touch responders.
  • Frictionless Commerce Evaluation for Judges: Mobile hackathons often suffer when judges cannot test paywalls due to missing StoreKit sandbox credentials. We resolved this by building a resilient Test Store fallback state machine in PurchaseService that provides instant entitlement toggling across both physical devices and web simulators.

🏆 Accomplishments That We're Proud Of

  • 29/29 Automated Unit Tests Passing (100% Deterministic Coverage): Comprehensive Jest test harness validating all critical system invariants across 7 suites (proof graph, matching engine, claim auditor, interview defense service, purchase service, proof pack service, and tutorial onboarding).
  • The 1-Tap Provenance Inspector: Achieving an effortless, tactile mobile interaction where tapping a resume bullet point instantly slides up the exact Git commit SHA and syntax extract.
  • Deep RevenueCat Value Alignment: Designing a monetization model where Pro genuinely provides ongoing utility (continuous repository sync and living career memory) rather than artificially gating basic features.

📚 What We Learned

  • Provenance is the Antidote to Cynicism: When AI generation is constrained to cite specific Git commit hashes and AST syntax nodes, applicant materials transform from dubious claims into verifiable engineering portfolios.
  • Deep Sponsor Architecture: Integrating RevenueCat through clean domain ports and reactive listeners makes mobile subscription engineering resilient, testable, and maintainable across development, testing, and production environments.

🚀 What's Next for Evident — Code-Grounded Career Intelligence

  • Automated GitHub App & Webhook Ingestion: Deploying an official GitHub App that automatically detects merged pull requests on main and continuously updates candidate evidence profiles in the background.
  • Public Recruiter Verification Gateways: Hosted, cryptographically signed web portfolios allowing engineering managers to inspect code provenance directly from a shared link without needing to clone repositories.
  • Multi-Platform Desktop Suite: Expanding Evident to native macOS and Linux environments for integrated IDE workflow support.

Built With

Share this project:

Updates

Submission history