🚀 Projects & Lessons Learned
Over the course of my career as a full-stack software engineer, I’ve worked on projects that combine modern web development with AI, real-time systems, and data-driven decision-making. Here’s a closer look at some of them — what I built, how I built it, and what I learned.
🧠 Udeffy – AI-Powered Personalized Learning Platform
Goal: Help learners master skills faster through adaptive learning paths.
What I Built:
- Backend in NestJS with a clean modular architecture
- AI lesson generation pipeline with credit-based access control
- Background processing with BullMQ and job event streaming via Server-Sent Events (SSE)
- Frontend in Next.js with rich UI using Tailwind CSS + Shadcn UI
Key Challenges & Solutions:
- Challenge: Generating diverse lesson formats (article, video, quiz, assignments) without blocking the main thread. Solution: Offloaded generation to BullMQ jobs and used event streaming for live updates.
- Challenge: Personalizing learning without overwhelming users. Solution: Used survey-based pre-tests to tailor learning content dynamically.
Outcome: A scalable platform that can handle multiple concurrent learners with individualized AI-driven content.
🏡 LinkAI – AI for Smarter Real Estate Decisions
Goal: Provide actionable insights for property buyers, investors, and agents.
What I Built:
- Property valuation and trend analysis engine
- Integration with real-time market data APIs
- Custom Copilot-style buyer agent powered by prompt engineering
- Output guidance per focus key (e.g., price, schools, crime) for relevant sections
Key Challenges & Solutions:
- Challenge: Providing accurate insights from volatile market data. Solution: Combined API data with AI summarization and strict business logic rules to ensure reliability.
- Challenge: Keeping AI answers on-topic. Solution: Designed structured prompts with guardrails to avoid hallucinations.
Outcome: An intelligent assistant that produces market insights users can trust.
📍 AI-Powered Map Discovery Tool
Goal: Help users find places that match specific personal preferences.
What I Built:
- Map-based search with conversational AI interface
- Criteria filtering (e.g., “restaurants with a sunset view and vegan options”)
- Favorites system for saving places
- Next.js frontend with real-time map rendering
Key Challenges & Solutions:
- Challenge: Translating vague user requests into actionable map queries. Solution: Used AI to parse natural language into structured location search criteria.
Outcome: A conversational location search that feels like talking to a local expert.
🔌 AI-Assisted Wiring Diagram Tool
Goal: Speed up wiring diagram creation for engineers and designers.
What I Built:
- AI layout optimization for spacing and alignment of components
- Assisted drag-and-drop UI for quick design
- Export options for various industry-standard formats
Key Challenges & Solutions:
- Challenge: Maintaining precision in AI-generated layouts. Solution: Implemented constraint-based algorithms that respect design rules.
Outcome: Reduced manual layout time significantly for technical teams.
📚 Lessons Learned Across Projects
- AI is only as good as the workflow around it — guardrails and validation are critical.
- Real-time features need strong architecture — event-driven systems scale better than polling.
- User experience is king — even the smartest AI fails if the UI is clunky.
- Scalable patterns matter — modular NestJS architecture + Next.js SSR is a winning combo.
💡 What’s Next
I’ll continue sharing deep dives into:
- AI integration in full-stack apps
- NestJS backend patterns
- Real-time event streaming
- Modern frontend performance tuning
If you want to follow along, keep an eye on my blog or connect with me:
- GitHub: [https://www.github.com/tunglamforwork]
- LinkedIn: [https://www.linkedin.com/in/ttlamdev/]
- Portfolio: [https://www.ttlam.dev]