DEV Community: Shyamli Khadse

Architect's Compass: Building a Data-Driven Trade-off Engine with AWS Kiro

Shyamli Khadse — Sat, 10 Jan 2026 15:26:02 +0000

Architect's Compass: Building a Data-Driven Trade-off Engine with AWS Kiro

In modern cloud architecture, "analysis paralysis" is a common bottleneck. Choosing between AWS Lambda and Amazon EKS (Kubernetes) isn't just about technical preference; it’s about weighing development velocity against operational cost and scalability.

I built Architect's Compass to solve this. Instead of providing a static answer, this tool evaluates user constraints to quantify trade-offs, helping developers choose the path that best aligns with their business goals.

The Core Challenge: Quantifying the "Choice"
Most comparison tools just list features. To help users actually choose, I implemented a Weighted Utility Engine. The engine takes six key dimensions—Time-to-Market, Cost, Maturity, Learning Curve, Scalability, and Maintainability—and adjusts their importance based on dynamic user input.

Accelerating Development with AWS Kiro
The most significant factor in building this tool was the Spec-Driven Development workflow enabled by AWS Kiro. By maintaining a /.kiro directory at the project root, I ensured the implementation never drifted from the design.

How the .kiro Directory Optimized My Workflow:
Requirements Synchronization: My requirements.md file contained 35 specific acceptance criteria. Kiro used these to audit my code, catching edge cases in the scoring logic before I even ran the first test.

Zero-Refactor Architecture: By defining interfaces in design.md before coding, the frontend components and the backend comparison engine were perfectly compatible on the first try.

Traceable Tasks: Following the roadmap in tasks.md allowed for a 30-40% reduction in total development time compared to traditional manual coding.

Key Results & Takeaways
The final application doesn't just name a winner; it surfaces a Trade-off Summary. If a user chooses Lambda for speed, the tool explicitly warns them about potential "Cold Starts" and "Vendor Lock-in," ensuring they are aware of the technical debt they are accepting.

Lessons for AWS Builders:
Specification is Foundation: Spending time in the /.kiro folder upfront saves hours of refactoring later.

Logic over Information: Tools should help users make decisions, not just consume data.

Modular Scoring: Designing pluggable scoring algorithms makes it easy to add new comparisons, like RDS vs. DynamoDB, in the future.

Project Resources
GitHub Repository: https://github.com/kshyamli/tradeoff-navigator/tree/main

Built with: React, TypeScript, Tailwind CSS, and AWS Kiro.

The Future of Hyper-Local AI

Shyamli Khadse — Sun, 28 Dec 2025 14:35:33 +0000

Building "Aamchi Guide" was more than just a coding exercise; it was an exploration into how Agentic Workflows can bridge the gap between global AI models and local cultural nuances. By utilizing Kiro’s context-injection capabilities, I was able to transform a generic chatbot into a street-smart Mumbai companion that understands not just the geography of the city, but its "vibe."

The true power of this project lies in its Complex Intent Orchestration. By teaching the AI to parse multiple needs—like hunger, location, and fatigue—simultaneously, we move closer to a future where AI feels less like a search engine and more like a helpful local friend. Whether it's guiding a tired traveler to a Marine Drive sunset or providing a Marathi lifeline in a medical emergency, "Aamchi Guide" proves that grounding AI in regional context is the key to building technology that truly serves India’s diverse population.

As we move forward in the AI for Bharat journey, the lesson is clear: Context is King. When we combine sophisticated logic with local heart, we create tools that don't just provide data—they provide solutions.

check out my project on github:https://github.com/kshyamli/aamchi-mumbai-guide/tree/main

🕹️ Retro Revival: Pitting Human vs. A* Search Intelligence

Shyamli Khadse — Sun, 21 Dec 2025 13:36:46 +0000

🚀 The Vision

For this week's challenge, I decided to breathe new life into the 1976 classic, Snake. The goal wasn't just to recreate the nostalgia of the Nokia era, but to introduce a Modern AI Twist: a competitive "Ghost" snake that uses high-level pathfinding logic to outsmart the player.

🧠 The "Modern Twist": Complex AI Logic
The heart of this project lies in ai.js. Unlike original arcade games that used simple "if-else" logic, my version implements the A (A-Star) Search Algorithm*.

How the AI Thinks:
The AI treats the game canvas as a grid of nodes. Every 100ms, it performs a search to find the shortest path to the food. It calculates the Manhattan Distance to prioritize moves:

                d = |x_1 - x_2| + |y_1 - y_2|

By assigning a "cost" of infinity to pixels occupied by snake bodies, the AI dynamically navigates around the player in real-time.

🛠️ How Kiro Accelerated My Build
The Kiro IDE was more than just a code editor; it was a lead engineer. Here is how it accelerated my development:

Algorithmic Scaffolding: Writing A* from scratch is error-prone. Kiro generated the initial class structure and heuristic functions based on my plan.md.

The "Straight Line" Fix: At one point, the AI snake would only move in a straight line. I shared the code with Kiro, and it identified that the target coordinate wasn't being refreshed inside the requestAnimationFrame loop.

Retro Styling: Kiro helped me write the CSS for the CRT Scanline Effect and the Windows 95 beveled buttons, giving it that authentic "Retro Revival" look.

📺 Visuals & Results
The final result is a seamless blend of 90s aesthetics and modern search theory.

UI: Windows 95 Teal Desktop.

Logic: Real-time adversarial pathfinding.

Proof of Work: All development logs are stored in my /.kiro directory.

"Kiro allowed me to focus on the high-level game design while it handled the complex matrix math required for the AI's brain."

🔗 Project Links
GitHub Repository:(https://github.com/kshyamli/Retro-Revival-AI-Enhanced-Snake)