GradNavi: designing and delivering an AI-powered career platform.

01 / PRODUCT DESIGN & IDENTITY

From wireframes and branding to a usable application

I created GradNavi's logo and brand identity and designed the broader low- and high-fidelity experience in Figma, including the student journeys and interface layouts. The visual system informed the public landing page, authentication screens, student navigation and learning/career workflows. This design authorship is my first-hand account; the connected Figma revision history was not independently available for this audit.

The original GradNavi Figma file contains the design source for the branding and application wireframes; its frames are linked, not recreated or substituted with illustrative mockups.

View GradNavi design workspace ↗  ·  GitHub implementation: #6, #76, #88.

02 / MY IMPLEMENTATION

Full-stack contributions across the student experience and AI services

My implementation work spanned the early frontend foundation, student workflows, supporting Django services, system integration, quality and later redesigns. These examples are tied to actual implementation PRs—not the original WBS task ownership.

Student experience and frontend engineering

• Foundation and onboarding: React/Vite routing, the original Home Page, login/registration and Student Profile interfaces, followed by authentication and profile API integration (#6, #8, #12, #14, #23).
• Career guidance: Career Recommendations, Career Readiness, Skill Gap Analysis, shared career selection, Career Roadmap, Learning Resources and Explore Careers interfaces (#44, #47, #50).
• Application and progress workflows: Resume and Cover Letter Builder interfaces and export/draft flows, Job Matching interface, Student Dashboard, account settings, global homepage/authentication redesign and responsive student-page improvements (#59, #61, #75, #76, #77, #88).

Backend, scoring and AI engineering

• Career intelligence: PostgreSQL configuration, reference data, weighted recommendation logic, readiness/gap services and composite scoring with semantic alignment (#1, #26, #27, #28, #45).
• AI and privacy: structured AI input/output contracts, reusable prompts, privacy allowlist and provider boundary; interview questions/feedback endpoints; external OpenAI provider integration and response validation (#36, #38, #65, #66).
• Service enhancements: learning roadmap/resource backend changes, Explore Careers evaluation API, deterministic job-description matching and additional backend work in the document-builder implementation (#46, #48, #49, #59, #60).

03 / PLANNING, QUALITY & RELEASE

Keeping an evolving capstone deliverable

Beyond feature implementation, I maintained extensive project-management and delivery work: plans and schedules, integration/testing trackers, meeting evidence, sprint reconciliation, defect follow-up and release preparation. My account is that I carried most of the practical coordination work despite the planned rotating-leader schedule. GitHub substantiates substantial documented management contributions; it cannot independently calculate individual effort percentages or every offline follow-up.

Representative records: API design #2, security architecture #3, schedule #18, meeting evidence #42, Sprint 5 planning #79, release #101 and closeout #103.

04 / COLLABORATIVE DELIVERY

What teammates implemented

Enamul121 — backend services

Enamul authored the authenticated registration/login APIs and student profile models/API (#9, #22), recommendation endpoint and initial learning-roadmap API (#29, #35), resume/cover-letter generation backend services (#37, #40), and administrator services and audit enforcement (#62, #70). Backend/database deployment configuration is recorded in #93.

Joyee8 — early design concepts, administration UI and deployment

Joyee contributed early Figma foundations and selected preliminary screen layouts (#32), and authored the original Interview Preparation interface and question/feedback flow (#63), a substantial administrative interface with protected routes and management screens (#64), and frontend deployment/asset optimisation work (#92, #94, #95).

This distinction matters: the backend and interface components were delivered collaboratively. The original WBS task allocations do not accurately describe every final implementation or the final design authorship; the early Figma concepts are distinct from the broader design work described above.

05 / IMPORTANT IMPLEMENTATION CORRECTION

Interview Preparation: original UI and later integration

Joyee's PR #63 contains the original Interview Preparation page, practice component and API service. That PR was closed without a direct merge. Later, my PR #78 merged the interview workflow with additional session persistence, validation and cross-feature changes. Comparing their added code reveals substantial identical UI and service lines.

The correct attribution is **original Interview Preparation UI: Joyee; subsequent integration, expansion and validation: Jerald**. Being the author of the merged integration PR is not evidence of having independently written all code in it.

06 / VERIFIED ENGINEERING ARCHITECTURE

How the shipped system fits together

The React student and administrator interfaces call Django REST Framework endpoints backed by PostgreSQL. Career and skill reference records support the scored recommendations, readiness and skill-gap services. Document and interview features use a separate AI service layer for structured prompt contracts, privacy-filtered context, provider integration and response validation.

• Deterministic fit and readiness: career and skill rules produce explainable component scores and a weighted combination; semantic embedding alignment contributes a distinct matching signal, rather than treating a text generator as the sole scoring authority. Evidence: #27, #28, #45.
• Generation boundaries: shared input/output contracts, prompts, safety rules and a privacy allowlist were developed before connecting the external text provider. The later OpenAI integration includes error handling and validation. Evidence: #36, #65, #66.
• Collaborative application delivery: Enamul's document-generation backend services (#37, #40) are used by the React resume and cover-letter builder interfaces implemented in Jerald's #59. Joyee's original interview interface (#63) was later integrated and extended in Jerald's #78.

Engineering lesson: keeping scored decision-support logic, AI-generated text and API contracts separate made it easier to validate behaviour, trace changes and coordinate frontend/backend work. Integration PR ownership is not the same as original code authorship.

07 / QUALITY & VALIDATION

What the repository records—and what it does not prove

Internal UAT

11 / 11 cases reported PASS

Production verification

21 / 21 cases reported PASS

Delivery record

Sprint 5 merged in PR #101

The repository's 13 October closeout summary attributes these verification outcomes to team execution reported by the project lead. Tests and production checks support the implementation record, but are not independent certification of AI accuracy, security or professional career-advice quality. Formal delivery acceptance was tracked separately, and final evidence reconciliation was noted as ongoing.

Source: GradNavi README and repository evidence and Sprint 5 PR #101.

08 / ENGINEERING OUTCOME

A capstone prototype with documented implementation and validation

Application

React + Django REST Framework

Storage and AI

PostgreSQL + OpenAI service layer

Evidence reviewed

106 GitHub pull requests

GradNavi's recorded implementation supports student career exploration and guidance workflows, AI-assisted drafts and interview practice, and administrator functions. The team carried out unit, integration, user acceptance and deployment-verification work. This is a student capstone prototype offering decision support, not independently validated professional career advice.

Explore the GradNavi team repository ↗

Attribution in this public case study is grounded in GitHub PRs, merged code and overlapping original implementations; it does not claim that PR author counts represent individual workload or exclusive ownership.