Proof Of Works~ Work that made real world impact

Context

Large open-source contribution events struggle with fair, scalable PR evaluation. Manual difficulty tagging (L1/L2/L3) is subjective, inconsistent, and vulnerable to bias.

Goal

Design and deploy a fully automated, deterministic PR scoring engine capable of evaluating thousands of pull requests across dozens of repositories without manual intervention.

Engineering highlights

  • ✨ Evaluated 14,500 Pull Requests

    Processed and deterministically scored over 14,500 merged PRs during the event without manual intervention.

  • ✨ Configured Across ~100 Repositories

    Deployed as a GitHub App installed across nearly 100 independent open-source projects, enabling centralized yet repository-isolated scoring.

  • ✨ Handled 2,500+ Contributors

    Supported large-scale participation with thousands of contributors submitting PRs concurrently across multiple repositories.

  • ✨ Deterministic & Auditable Scoring (No AI)

    Implemented fully rule-based scoring using objective signals (LOC, files changed, density, test additions, structural impact), ensuring transparency and zero hidden bias.

  • ✨ Idempotent & Replay-Safe Processing

    Ensured each PR is scored exactly once, preventing duplicate webhook deliveries or retries from inflating contributor scores.

Impact

  • 100+ repositories integrated.
  • 2,500+ contributors evaluated.
  • 14,500 merged PRs processed and scored.
  • Zero manual difficulty assignments required.

Outcomes

Transformed PR evaluation from a subjective, manual bottleneck into a deterministic, scalable, and fully auditable system — restoring fairness and trust across a large open-source event.

Developer Tooling • Open Source Infrastructure • Event Systems

ECWoC Sentinel – Fair & Transparent PR Scoring System

Design and deploy a fully automated, deterministic PR scoring engine capable of evaluating thousands of pull requests across dozens of repositories without manual intervention.

Architecture

Event-driven backend system using GitHub webhooks, Redis-backed job queues, and a deterministic rule engine to score merged PRs across 100+ repositories.

Node.jsExpress.jsBullMQRedisMongoDB AtlasGitHub App
See where it's in use
View on

Context

Modern AI-driven products require aggregating deeply personal, behavior-based user data to deliver personalization and intelligent insights.

Goal

Design a modular, privacy-by-default context layer that consolidates sensitive user data, enforces consent-gated decryption, and safely exposes sanitized aggregated context to AI agents and personalization systems.

Engineering highlights

  • ✨ Selective Field-Level Encryption

    Implemented AES-256-CBC encryption for narrative-sensitive fields (e.g., notes) using ivHex:cipherHex format, ensuring raw sensitive data is never stored in plaintext.

  • ✨ Consent-Gated Decryption Model

    Encrypted fields are entirely omitted unless explicit AI consent is granted, preventing accidental exposure to AI agents.

  • ✨ Idempotent Context Bootstrap

    Built /api/context/init to safely create only missing documents, enabling reliable onboarding and re-initialization without duplication.

  • ✨ Extensible Context Pattern

    Defined a clean model + controller + route pattern so new context types can be added without touching aggregation core logic.

Impact

  • 100% of sensitive narrative fields encrypted at rest.
  • Zero sensitive fields exposed to AI without explicit consent.
  • Single aggregation endpoint powering personalization & AI features.

Outcomes

Created a reusable privacy foundation that allows AI-powered features to ship safely without re-solving encryption, validation, or consent logic for every feature.

Backend systems • Privacy engineering • AI infrastructure

Privacy-First Context Layer

Design a modular, privacy-by-default context layer that consolidates sensitive user data, enforces consent-gated decryption, and safely exposes sanitized aggregated context to AI agents and personalization systems.

Architecture

Modular context collections with selective field-level encryption, Zod-validated controllers, consent-gated aggregation, and a clean service-layer orchestrator.

Node.jsExpress.jsMongoDBZodAES-256-CBCJWT (HTTP-only cookies)
See where it's in use
View on

Context

Polaris ’26 required a secure and scalable digital system to manage high-volume student registrations, OTP verification, QR issuance, and multi-stage event entry.

Goal

Build a deterministic, race-safe, and production-ready RSVP and gate system capable of handling thousands of registrations and emails within a short event window.

Engineering highlights

  • ✨ Handled 1,800+ Registrations in 72 Hours

    System processed 1,800 total registrations, with ~1,500 successfully verified and acknowledged within a 2–3 day event window.

  • ✨ Processed ~2,000 Transactional Emails Reliably

    OTP and confirmation emails were delivered using a rotating Gmail infrastructure without hitting provider hard-block limits.

  • ✨ Race-Safe Multi-Stage Gate Enforcement

    Used conditional atomic updates (findOneAndUpdate with scannedCount < maxScans) to prevent duplicate or concurrent over-entry.

  • ✨ Zero Plaintext OTP Storage

    Stored only bcrypt-hashed OTPs with expiration windows, preventing replay and credential exposure risks.

Impact

  • 1,800 total registrations processed in ~72 hours.
  • ~1,500 verified attendees acknowledged successfully.
  • ~2,000 transactional emails delivered across rotating Gmail accounts.
  • Zero over-scan incidents due to atomic scan constraints.

Outcomes

Replaced manual verification with a deterministic, scalable gate system capable of handling burst traffic and real-world operational pressure during a large campus event.

Full-stack systems • Event infrastructure • Authentication & security

Polaris ’26 – RSVP, Email & Multi-Stage QR Entry System

Build a deterministic, race-safe, and production-ready RSVP and gate system capable of handling thousands of registrations and emails within a short event window.

Architecture

Stateless Next.js APIs backed by MongoDB, OTP-gated QR activation, JWT-based committee auth, and atomic scan updates enforcing bounded multi-stage entry.

Next.js (App Router)Node.jsMongoDB + MongooseJWTbcryptNodemailer
See where it's in use
View on
Home

GUNJANGHATE.

I'm Gunjan - a Web2/Web3 developer, knows both frontend and backend & a good problem solver. Thanks for checking out my site!

Nagpur, Maharashtra, India
© 2025 Gunjan Ghate ~ All Rights Reserved

General

Feedback~ Drop a Thought, Spark a Change