Inspiration
Across Bangladesh, millions of students, job seekers, and aspiring freelancers navigate three disconnected worlds: they study NCTB and university textbooks without personalized guidance, face high-stakes BCS, banking, and software engineering viva boards without realistic practice, and lose global freelance contracts on Upwork and Fiverr due to English communication barriers. We built SkillBridge 2.0 to bridge this gap—uniting bilingual education, career readiness, and global freelancing into one multi-agent AI ecosystem that works seamlessly in বাংলা (Bangla), English, and Banglish.
What it does
SkillBridge 2.0 unifies three flagship workspaces with a Digital University and Career Operating System:Module A — PorikhaGPT (AI Education & Admission Tutor):Uses Gemini Vision (inlineData) to inspect handwritten math and physics derivations (for example, differentiating $y = e^{\sin(x^2)}$ to obtain $\frac{dy}{dx} = 2x\cos(x^2)e^{\sin(x^2)}$) and explain step-by-step solutions, board shortcuts, and common student mistakes.Indexes uploaded NCTB and university PDFs using deterministic RAG chunking (chunkDocument with $900$-character chunks and $150$-character overlap) and top-$K$ retrieval (retrieveRelevantChunks).Generates adaptive 4-question quizzes that only promote a skill from AI-Inferred to Demonstrated when the learner achieves a verified score of:$$\text{Score} = \frac{\text{Correct Answers}}{\text{Total Questions}} \times 100\% \ge 70\%$$Module B — ChakriPrep AI (Bangladesh Job & Viva Simulator):Benchmarks CVs against Bangladesh Software Engineering, BCS Cadre, Bangladesh Bank (AD), and private-sector job descriptions with a zero-fabrication guardrail that never invents credentials.Conducts live AI Mock Vivas in Voice (Web Speech API STT/TTS), Text, and Timed ($120\text{s}$) modes, scoring candidates across an 8-criteria rubric and generating a 3-week preparation roadmap.Module C — Bhasha Bridge (AI Freelancing & Client Communication Studio):Executes a 6-step workflow (Bangla/Banglish Input → Intent Detection → Requirement Analysis → Proposal Generation → Scope Estimation → Polished English Output).Computes transparent 4-milestone project budgets in both USD ($\$$) and BDT ($\text{৳}$) at a configurable conversion rate ($\text{৳}120/\text{USD}$) alongside an interactive 10-stage client negotiation simulator.Digital University, Community 2.0 & Career OS:Connects an 18-node interactive Knowledge Graph (with $<75\%$ prerequisite gap alerts), Challenge Arena, Collaborative Project Lab 2.0 (requiring GitHub or live URL proof), AI Research Assistant, Real-Time Community Hub, and a 43+ verified Bangladesh Resource Directory.
How we built it
Frontend: Engineered with React 19, TypeScript, Tailwind CSS 4, Vite, and Framer Motion (motion), packaged as an installable Progressive Web App (vite-plugin-pwa) with full i18next bilingual (bn / en) localization.
Backend & 9-Stage Agentic Orchestrator: Built an Express.js API gateway (server.ts) that keeps GEMINI_API_KEY strictly server-side and routes requests through a 9-Stage Control Loop (Intent Detection → Profile Context → Task Planning → Agent Selection → Execution → Verification) across 17 specialist domain agents powered by @google/genai (Google Gemini Flash).
Database & Security: Powered by Firebase Authentication, Firebase Admin, and Cloud Firestore with owner-isolated collections enforced by firestore.rules and verified via automated test suites (orchestratorLogic.test.ts and firestore.rules.test.ts).
Challenges we ran into
Eliminating Hallucinations in CVs, Research & Skill Claims: Standard LLM wrappers frequently invent job experience, fabricate academic DOIs, or mark unverified skills as mastered. We solved this by separating AI-Inferred skills from quiz-verified Demonstrated skills, enforcing URL evidence guards in Project Lab 2.0, and restricting voice viva feedback to observable pacing and clarity rather than fabricated psychological claims.
Bilingual & Banglish Nuance Across Voice and Text: Bangladeshi learners naturally mix Bangla script, phonetic Banglish, and technical English terms. Designing prompts and Speech-to-Text workflows that preserve technical terminology while explaining concepts naturally in Bangla required dedicated language-mode routing across all 17 agents.
Accomplishments that we're proud of
Designing a Unified Cross-Module Skill Graph where solving a calculus or coding quiz in PorikhaGPT automatically updates the learner's 7-Dimension Career Readiness Assessment in ChakriPrep AI and populates their verifiable portfolio in Bhasha Bridge.
Combining deterministic TypeScript application logic (RAG chunking, prerequisite graph validation, and milestone budget math) with structured JSON generation from Google Gemini.
What we learned
Building trustworthy AI for education and employability requires pairing generative models with deterministic verification layers, explicit human-in-the-loop review (such as side-by-side diffs in Document Studio), and strict owner-isolated cloud security rules.
What's next for SkillBridge 2.0
Integrating live bKash and SSLCommerz merchant checkout for our accessible student subscription tier ($\text{৳}30\text{–}\text{৳}490\text{ BDT}$).Expanding offline PWA caching for rural low-bandwidth learners across Bangladesh and partnering with university career and programming clubs.
Built With
- cloud-firestore
- css3
- express.js
- firebase
- framer-motion
- git
- google-gemini
- html5
- i18next
- javascript
- json
- node.js
- nodemailer
- pwa
- react
- rest-api
- tailwindcss
- typescript
- vite
- web-speech-api
