Full-stack developer moving into AI engineering through developer tooling, LLM context management, agent workflows, and evaluation systems.
I work primarily with TypeScript, JavaScript, and Python. My recent work focuses on building web products and exploring a practical problem in AI-assisted development: turning large, messy codebases into context that language models can use effectively.
A CLI that packages codebases into deterministic, AI-ready Markdown context files.
It supports:
- 32k, 64k, and 128k token-budget splitting
- Git-aware modes for changed, staged, stashed, and branch-diff files
- Skeleton extraction for lightweight JavaScript and TypeScript structure
.gitignoreand.kontxtignorerules- Deterministic file ordering and output
Current npm release: 0.1.4
Install it globally:
npm install -g kontxt-cli# Package the full repository into one context file
kontxt -e
# Split a large repository into files with a 64k token budget
kontxt -e --64k
# Package only changed, staged, and untracked files
kontxt -e --changed --32k
# Package only files prepared for the next commit
kontxt -e --staged --32k
# Package work on the current branch since it diverged from main
kontxt -e --since main --64k
# Package the latest Git stash without modifying the worktree
kontxt -e --stash --32k
# Extract lightweight code structure instead of full JS/TS contents
kontxt -e --skeleton --32k
# Print the repository tree without creating context files
kontxt -tGenerated files are written under .kontxt/.
A course marketplace MVP built with Next.js, TypeScript, Prisma, PostgreSQL, Clerk, and Stripe. It includes instructor course management, role-aware access, checkout, webhook-driven enrollment, and student learning views.
A Manifest V3 Chrome extension for YouTube watch-time analytics and distraction control. It tracks creator-level usage, handles YouTube's client-side navigation, stores analytics locally, and injects focus controls into the existing interface.
An early-stage technical interview readiness product being built around curated questions, controlled test attempts, and post-attempt AI evaluation.
The current implementation includes Google authentication, database integration, basic profile creation, placeholder onboarding, and foundational UI components. The test engine, question-bank workflow, evaluation pipeline, reports, and study plans are still planned work.
Planned product scope
- Google authentication and developer onboarding
- Education, experience, target-role, and technology-stack profiles
- Test selection by topic, experience tier, target role, and question count
- Randomized tests generated from a curated question bank
- One-question-at-a-time answering with saved progress
- Previous-attempt and score history
- Imported-question classification by category, tag, difficulty, and tier
- Complete-attempt evaluation after submission
- Overall and tier-readiness scores
- Per-question feedback and knowledge-gap analysis
- Communication feedback, study plans, and interviewer-style notes
- Recommended next tests and difficulty tiers
- CSV question import
- Curated question-bank management
- Manual correction of AI classifications
- Category, tag, tier, and status filtering
- Question activation and deactivation
- No video proctoring
- No advanced cheating detection
- No recruiter dashboard
- No live AI help during tests
- No fully AI-generated runtime question bank
TypeScript JavaScript Python Node.js Next.js React PostgreSQL Prisma Drizzle ORM
Exploring agent workflows, retrieval, evaluation, and observability through small projects using LangGraph, promptfoo, and Langfuse.
Open to frontend, full-stack, developer-tooling, and AI-product internships.