karthik.dev

Technical Blog

Engineering Notes

Writing on AI engineering, system design, and building in public — the AI Recruiter Assistant can answer questions about these posts too.

AI Engineering

TraceGraph AI: Building a Knowledge Graph for LLM Telemetry Diagnostics

How I built TraceGraph AI — an evidence-first platform that constructs a Neo4j knowledge graph from three bounded real-world inputs (product docs, Playwright crawl, GitHub PR diffs) to surface root causes in LLM architecture failures.

4 min readRead
AI Engineering

aiskills: Building a Reusable AI Engineering Skills Library

How I built aiskills — an open-source, tool-agnostic library of reusable AI engineering skills, RAG pipeline templates, LLM evaluation harnesses, and agent workflow blueprints for AI coding agents across Cursor, Claude, Windsurf, and VS Code.

3 min readRead
MCP

Building a Production MCP Server from Scratch

How I designed, built, and published a Model Context Protocol (MCP) server exposing 10 tools that power 167+ AI agents in Career-Agents — covering tool schema design, error handling, and client integration with Claude Desktop, Cursor, and Windsurf.

4 min readRead
NVIDIA

NVIDIA Agent Doctor: Building a GPU AI Environment Diagnostics CLI

How I built NVIDIA Agent Doctor — an open-source CLI that automates GPU, CUDA, and PyTorch compatibility checks, security audits, and bounded benchmarking for NVIDIA AI environments, with MCP server integration for agent-driven diagnostics.

4 min readRead
RAG

GP Support Agent: Building a Deterministic RAG + LangGraph Agent for Healthcare

How I built a production-grade, deterministic RAG + agent architecture using LangChain and LangGraph for a GP/healthcare-reception knowledge base — with an integrated evaluation harness to measure groundedness, faithfulness, and coverage.

4 min readRead
AI Watermarking

ClaudeMark: Building an Open-Source AI Watermark & Provenance Forensics Platform

How I built ClaudeMark — a local-first Python platform combining statistical detection, Unicode steganography analysis, C2PA/metadata inspection, SARIF/CI integration, and forensic diffing to investigate AI-generated content without external cloud services.

3 min readRead
AI Agents

Building a Multi-Turn Tool-Calling WhatsApp AI Agent with Claude

How I built Cody — a production WhatsApp AI sales agent using Flask and the Claude API with multi-turn tool-calling, Hindi/Hinglish language mirroring, and a live monitoring dashboard for real-time production visibility.

5 min readRead
Android

GPU Insight AI: Building an Android GPU Monitor with Gemini AI Diagnostics

How I built GPU Insight AI — an open-source Android app using Kotlin, Jetpack Compose, Room, and Gemini AI to monitor GPU hardware telemetry in real time and provide AI-powered diagnostics for performance bottlenecks.

4 min readRead
Engineering

How I Built a Real-Time LeetCode Stats Dashboard with 2-Minute Polling

Behind the scenes of fetching live LeetCode data server-side, avoiding CORS, caching intelligently, and building a UI that tells you exactly when data is fresh.

7 min readRead
AI Engineering

LearnOS AI: Building a Personalized AI Learning Operating System

How I built LearnOS AI — an AI-powered Learning OS that transforms a learner's background and target role into a personalized roadmap, curated projects, AI workspaces, and interview preparation tracks using OpenAI and Next.js.

5 min readRead
Enterprise AI

Kaegis AI: Designing an Enterprise AI Operating System

How I designed Kaegis AI — an open Enterprise AI Operating System for building, deploying, governing, and scaling enterprise AI applications. Covers the governance gap, audit trail design, model access control, and agentic policy enforcement.

3 min readRead
RAG

CodeRAG: AST-Parsed Codebase RAG with Hybrid Search and Precise Citations

How I built CodeRAG — a production-ready RAG engine for codebases that uses Tree-sitter AST parsing to chunk at semantic boundaries, combines dense + sparse hybrid search, and generates responses with precise file and line number citations.

4 min readRead
DSA

Why Every Developer Should Solve 50 Hard LeetCode Problems

Not because interviews ask for them — but because hard problems rewire how you think about constraints, edge cases, and trade-offs in production systems.

5 min readRead
AI Engineering

JobPilot AI: Building a Full-Stack AI Job Copilot with BullMQ and Gemini

How I built JobPilot AI — a production-ready personal AI job copilot using Next.js 15, BullMQ-backed background job processing, Redis caching, and Gemini AI for job match scoring, tailored application generation, and interview preparation.

5 min readRead
Backend Engineering

Building a Payment Gateway from Scratch: What the Tutorials Don't Tell You

Idempotency keys, dead-letter queues, row-level locking, webhook retries — the unglamorous infrastructure that makes real money move correctly.

9 min readRead
AI Agents

Kestrel: Building an Autonomous Software Engineering Platform

How I built Kestrel — an open-source autonomous software engineering platform where AI agents understand repositories, analyze GitHub issues, write code, run tests, and open pull requests through a distributed BullMQ orchestration engine.

4 min readRead
AI Engineering

AdIntel AI: Building an AI Campaign Intelligence Platform for Media Buyers

How I built AdIntel AI — a campaign intelligence platform that consolidates Meta, Google, TikTok, and Taboola ad data into one view, using OpenAI to generate root-cause analysis and executive-ready performance briefs.

5 min readRead
AI Engineering

CareerByte AI: Building an Open-Source AI Career Copilot with Gemini and Prisma

How I built CareerByte AI — an open-source AI career copilot unifying job discovery, ATS resume optimization, application tracking, and interview preparation into a single Next.js + Prisma + Gemini AI workflow.

4 min readRead
SEO

AuditForge AI: Building a Five-Dimension Website Auditor with Gemini

How I built AuditForge AI — a platform that audits any website across SEO, performance, accessibility, security, and AI crawler visibility, using Gemini AI to translate raw findings into prioritized, plain-language recommendations.

4 min readRead
Frontend Engineering

Scroll-Driven Animations with Canvas: How This Portfolio's Background Works

300 JPEG frames, a requestAnimationFrame loop, lerp-based smoothing, and a DPI-aware cover-fit algorithm — the full technical breakdown.

8 min readRead
Multi-Agent

Multi-Agent Research Assistant: Planner, Search, Synthesis, and Writer

How I built a four-agent autonomous research pipeline — Planner decomposes queries, Search fetches real-time results via Tavily, Synthesis extracts claims and evidence, Writer produces a structured report with citations — using TypeScript and the OpenAI API.

5 min readRead
Backend Engineering

Redis Caching Patterns That Actually Work in Production

Cache-aside, write-through, TTL strategies, and the cache invalidation problems that will bite you — from real experience building CodeMyFYP Academy.

7 min readRead
RAG

RAG Chat Engine: Powering the Portfolio AI Chatbot with Zero External Vector DB

How I built the production RAG Chat Engine powering this portfolio's AI chatbot — a full Retrieval-Augmented Generation pipeline that embeds a knowledge corpus, retrieves by cosine similarity, and generates grounded answers using Gemini, with no external vector database.

6 min readRead
SaaS

NEXORA: Building a Multi-Role SaaS Platform with OpenAI and 80%+ Test Coverage

How I built NEXORA — a scalable multi-role SaaS platform combining JWT authentication, RBAC, OpenAI content generation, real-time analytics, and a fully automated CI/CD pipeline with 80%+ unit test coverage using Jest and GitHub Actions.

5 min readRead
SaaS

Building CodeMyFYP Academy: Multi-Tenant EdTech SaaS from Zero to AICTE

How I designed and built CodeMyFYP Academy — a full-scale multi-tenant EdTech SaaS platform serving students, educators, recruiters, and administrators across VTU, Bangalore University, and Mysore University, officially listed on the AICTE National Internship Portal.

5 min readRead
Frontend Engineering

Next.js App Router: Patterns I Use on Every Project

Server components, streaming, route groups, parallel routes, and the mental model shift that makes App Router click — from someone who's shipped multiple production apps with it.

8 min readRead
Engineering

From Founder to Engineer: What Building CodeMyFYP Taught Me About Systems

Starting a company at university and building the platform from scratch — the technical decisions, the mistakes, and the engineering instincts that came from owning every layer.

7 min readRead
DSA

DSA Algorithm Visualizer: Animating 20+ Algorithms Step by Step

How I built an interactive DSA Algorithm Visualizer using React, TypeScript, Canvas API, and Framer Motion — animating sorting algorithms, graph traversals, tree operations, and DP tables with speed controls and synchronized pseudocode highlighting.

5 min readRead
Backend Engineering

URL Shortener with Analytics: Sub-10ms Redirects and Async Click Processing

How I built a production-grade URL shortener with sub-10ms redirect latency using Redis caching, decoupled async click analytics with background processing, geographic IP resolution, and Docker-based deployment.

5 min readRead
DSA

Dynamic Programming: The Patterns Behind the Problems

After solving 60+ DP problems, I've found they mostly fall into five pattern families. Recognizing the family is 80% of solving the problem.

8 min readRead
Engineering

TypeScript Patterns I Use Every Day (And Why They Matter)

Discriminated unions, const assertions, template literal types, satisfies, and the pattern that finally made me stop fighting the type system.

6 min readRead
DSA

Graph Algorithms That Show Up Everywhere (And How to Spot Them)

BFS, DFS, Dijkstra, Union-Find, and topological sort — the five graph primitives that underlie hundreds of LeetCode problems and real system design challenges.

7 min readRead
AI Engineering

Prompt Engineering for Reliable AI Agents: What I Learned Building Career Agents

System prompts, JSON output schemas, temperature settings, chain-of-thought triggers, and the evaluation harness that tells you whether your prompts actually work.

7 min readRead
Backend Engineering

PostgreSQL Indexing: The Missing Guide for Application Developers

B-tree, partial, composite, and covering indexes — when to use each, how to verify they're being used, and the common mistakes that nullify your indexes entirely.

8 min readRead
DevOps

CI/CD with GitHub Actions: From Zero to Production-Ready Pipeline

How I set up automated lint, test, build, Docker, and deployment pipelines for my projects — the exact workflow files and the reasoning behind each step.

7 min readRead
AI Engineering

Designing Multi-Agent Systems That Don't Fall Apart in Production

Lessons from building Career Agents — why a single orchestrator, strict JSON contracts, and shared session memory matter more than clever prompts.

6 min readRead
Backend Engineering

JWT Authentication Done Right: Stateless Auth Without the Footguns

Access tokens, refresh tokens, secure storage, rotation strategies, and the common implementation mistakes that make JWT-based auth insecure — with fixes.

7 min readRead
Frontend Engineering

How I Optimized a Next.js App's Core Web Vitals from 60 to 95

LCP, CLS, FID — the actual changes (image optimization, font loading, code splitting, ISR) that moved the needle on real performance metrics.

6 min readRead
System Design

Five Roles, One Codebase: RBAC Lessons from a Multi-Tenant EdTech Platform

How CodeMyFYP Academy enforces data isolation across Students, Mentors, Recruiters, Faculty, and Administrators without duplicating business logic per role.

5 min readRead
SaaS Building

GTM Engineering: Why I Ship an llms.txt Next to My Sitemap

Technical SEO used to mean robots.txt and sitemaps. Now it also means making your site legible to AI crawlers and answer engines.

4 min readRead