DEV Community: Nabil Thange

I built an AI tutor with persistent memory in 6 hours — here's how

Nabil Thange — Fri, 20 Mar 2026 15:20:19 +0000

I built an AI tutor with persistent memory in 6 hours — here's how

Every study app I've used has the same problem.

Open it on Monday. It helps you. Close it.

Open it on Thursday. It has no idea who you are.

Same generic suggestions. Same random quiz topics. No memory of the fact that you failed recursion questions three times this week.

So for this hackathon — theme: AI Agents That Learn Using Hindsight — we built Sage. An AI tutor that actually remembers you.

What Sage does

You upload your syllabus PDF. Sage reads it, extracts every subject and chapter, and asks you what you already know. Then it generates a personalised study plan built around your actual weak areas — not a template.

Every time you:

Take a quiz and get something wrong
Ask the AI to re-explain a concept
Tell the planner you have an exam coming up
Finish a Pomodoro session

...Sage writes that to memory. Forever.

Come back three days later? The AI opens with:

"Welcome back. Last session you struggled with VSWR. Your IOT exam is in 4 days. Want to pick up from there?"

That's not a chat history trick. That's a persistent memory system.

The tech that makes it possible — Hindsight

The memory layer is powered by Hindsight by Vectorize.

Hindsight isn't just a key-value store. When you call retain(), it uses an LLM internally to extract facts, entities, and temporal data — then classifies them into 4 memory types:

World — objective facts ("IOT exam is March 28")
Experiences — what actually happened ("scored 4/10 on quiz, failed VSWR")
Observations — patterns it auto-synthesises ("user consistently fails application questions")
Opinions — beliefs with confidence scores ("needs more practice: 0.91 confidence")

And reflect() reasons across all of these to generate insights like:

"Strong in theory. Consistently weak in numerical and application problems."

This is what makes Sage feel like a real tutor — not just a chatbot.

Architecture

User → Next.js frontend
         ↓
/api/planner/chat   → recall() from Hindsight → Groq with memory context
/api/mentor/chat    → recall() + syllabus text → Groq → personalised teaching
/api/memory/retain  → Hindsight retain() on every meaningful interaction
/api/memory/reflect → Hindsight reflect() for insights + Memory Panel

Stack:

Next.js 14 (App Router)
Groq — qwen3-32b, fast and free tier
Hindsight Cloud — persistent per-user memory banks
shadcn/ui + Tailwind — dark-themed component library
pdf-parse — server-side syllabus extraction

The Memory Panel — making invisible AI visible

The coolest thing we built wasn't the chat. It was the 🧠 Memory Panel.

One click and you see everything the AI knows about you — grouped by type, with confidence scores on opinions, and a live reflect() summary at the top:

🌍 World
   BTech CSE Sem 3. IOT exam March 28.

🎯 Experiences
   Scored 4/10 on IOT quiz. Failed: VSWR, antenna gain.

🔍 Observations (auto-synthesised)
   Fails application questions consistently across 3 sessions.

💭 Opinions
   Needs IOT application practice — 0.91 confidence

This screen alone makes it obvious that the AI is actually learning — not just replying to prompts.

What we learned

1. Hindsight's reflect() is the killer feature. It's not retrieval — it's synthesis. The AI forming its own opinions about a student's learning patterns is something no RAG pipeline gives you.

2. Build memory-first, UI second. We wired Hindsight in the first hour. Every feature we built after that automatically had memory context. If you add memory last, you have to retrofit everything.

3. The "return visit" moment is your demo. Log out, log back in, watch the AI greet you with context from days ago. That's the moment that makes people go "wait, this is actually different."

What's next for Sage

NotebookLM MCP integration for Studio outputs (audio overview, infographic, slide deck)
Real Supabase persistence for test history and proficiency tracking
Smarter pre-test adaptive questioning

Built in 6 hours at a hackathon. Theme: AI Agents That Learn Using Hindsight.

If you're building anything with persistent AI memory — check out Hindsight. It's genuinely different from standard RAG.

Code dropping soon. Follow for updates.

🚀 Building Vyx: An AI Content Engine with the Kiro IDE

Nabil Thange — Fri, 05 Dec 2025 12:00:24 +0000

🚀 Building Vyx: An AI Content Engine with the Kiro IDE

By Nabil Salim Thange | Portfolio: http://nabil-thange.vercel.app/

The content creation world is drowning in manual labor. We realized that even with dozens of AI tools, creators were spending 15-20 hours a week just repurposing a single YouTube video for different platforms.

That’s why we built Vyx, an AI engine that converts any video URL into a complete, platform-optimized content package (blogs, social posts, viral clips) in under 10 minutes.

This project was complex, requiring the orchestration of multiple asynchronous microservices (Next.js, Groq, N8N, FFmpeg). But thanks to the Kiro IDE, we handled this complexity and drastically reduced our development time.

What is Vyx? The Solution to the Content Bottleneck

Vyx is a dual-pipeline system designed for creators and marketers:

Content Repurposing Pipeline (Pipeline A): Takes the YouTube transcript and runs it through a 5-step AI pipeline (Llama 3.3 70B) to generate a complete suite of content:
- Optimized blog posts (800-1200 words).
- Platform-specific social posts (LinkedIn, X, Instagram) with custom aspect ratio images.
- Content is scored for Virality and Usefulness.
Viral Clip Generation Pipeline (Pipeline B): Uses FFmpeg and yt-dlp to download the video, identifies the top 3-5 viral moments using AI, crops them to the 9:16 vertical format, and burns in subtitles.

The challenge wasn't just the AI; it was managing the infrastructure for instant content, complex processing, and multi-service orchestration.

🛠️ How Kiro Revolutionized Our Development Workflow

Developing Vyx on the Kiro IDE was essential for two primary reasons: enforcing structure and enabling complex multimedia processing.

1. Spec-Driven Development Eliminated Vibe Coding

The project’s architecture, managing multiple asynchronous service integrations (Next.js for the frontend/API, N8N for orchestration, Groq for AI logic, Pollinations.ai for instant images), was intricate.

Instead of getting lost in "vibe coding," Kiro allowed us to immediately transition from our Implementation Plan and PRD into concrete, working code.

Clarity from the Start: By defining the input and output for each module in the spec, Kiro was able to build a reliable N8N workflow structure for the backend microservices, which drastically reduced integration bugs.
Focus on Logic: Kiro handled all the boilerplate for the Next.js/Tailwind setup, allowing us to spend 95% of our time focusing on optimizing the 5-step AI pipeline and refining the viral detection logic.

2. The MCP Unlocked Multimedia Capabilities

The most critical feature—the Viral Clip Generation Pipeline (B)—relied heavily on non-standard binary tools: yt-dlp for downloading and FFmpeg for video manipulation.

This is where the Master Control Program (MCP) feature of Kiro was a game-changer:

Native Tool Integration: We extended Kiro’s capabilities to understand and orchestrate yt-dlp and FFmpeg commands.
Complex Workflow Solved: Kiro was able to stitch together a workflow that downloaded the video, identified timestamps via the AI, and then reliably executed the necessary FFmpeg commands to crop the video to 9:16 and burn in subtitles with precise timing.

Without the MCP, building a reliable, automated video processing pipeline would have been nearly impossible or prohibitively time-consuming.

Conclusion

Vyx is proof that AI orchestration, powered by the right development environment, can solve real, multi-dimensional business problems. The structural enforcement of Spec-Driven Development and the power of the MCP allowed us to build a complete, production-ready AI content engine in record time.

Kiro didn't just write code; it structured a complex technical project, enabling us (Nabil Salim Thange and team) to successfully launch a full-stack solution.

If you’re building a complex project with multiple integrations, I highly recommend exploring how Kiro IDE can streamline your workflow.

DEV Community: Nabil Thange

I built an AI tutor with persistent memory in 6 hours — here's how

I built an AI tutor with persistent memory in 6 hours — here's how

What Sage does

The tech that makes it possible — Hindsight

Architecture

The Memory Panel — making invisible AI visible

What we learned

What's next for Sage

🚀 Building Vyx: An AI Content Engine with the Kiro IDE

🚀 Building Vyx: An AI Content Engine with the Kiro IDE

What is Vyx? The Solution to the Content Bottleneck

🛠️ How Kiro Revolutionized Our Development Workflow

1. Spec-Driven Development Eliminated Vibe Coding

2. The MCP Unlocked Multimedia Capabilities

Conclusion

kiro