Hi, I'm Ayan.
A Full-Stack Developer originating structured memory architectures, graph-based navigation engines, and production-ready AI tooling. Reducing friction through novel abstraction systems.
Tech Stack
Selected Works
Supercharge
"Standard AI prompts are redundant and difficult to manage across multi-turn sessions."
- ▹Designed Personal Model Language (PML) to structure AI memory into modular, reusable injection blocks.
- ▹Built a web interface enabling dynamic AI workflows via BYOK (Bring Your Own Key) without backend overhead.
- ▹Architected a scalable context abstraction layer improving consistency and rebuild time.
CLI-AI
"Most AI tools require complex setup (installing CLIs, runtimes, configuring API keys) before generating code."
- ▹Turns AI code generation into a single curl download command.
- ▹Eliminates client-side setup by handling language inference and prompt execution entirely on the server.
- ▹Ensures universal accessibility across fresh laptops, shared lab computers, or remote SSH environments.
Steganography
"Secure communication flags itself as encrypted data; steganography hides the existence of the message."
- ▹Developed a system to securely embed encrypted messages within standard image files.
- ▹Ensured zero visual degradation to the host images, escaping standard detection methods.
- ▹Implemented multi-layered security protocols combining encryption and steganography.
Experience
AI/ML Lead
Delivered a fully functional offline indoor navigation engine within hackathon time constraints, enabling route computation across multi-floor buildings with zero GPS dependency.
Reduced manual mapping effort by ~60% by designing an ML pipeline that automatically parsed floor plan images into navigable graph nodes.
Implemented A* pathfinding on a graph-converted floor plan, achieving accurate multi-step route resolution across complex campus layouts.
Shipped a full PWA prototype supporting low-connectivity environments in collaboration with a cross-functional team.
Solo Developer
Designed PML (Personal Model Language), a custom context encoding system that reduced prompt redundancy by structuring AI memory into modular, reusable injection blocks.
Built a React + Vite web interface enabling dynamic AI workflows via user-provided API keys, supporting full context customization without backend overhead.
Architected a scalable context abstraction layer that improved AI response consistency and cut context rebuild time per session.
Implemented a structured parsing engine for contextual retrieval, enabling deterministic memory injection across multi-turn AI conversations.
Solo Developer
Built a terminal-based AI assistant that unified developer command workflows into a single abstraction layer, reducing context-switching overhead for common tasks.
Designed a structured I/O pipeline to minimize response latency and enforce execution clarity across multi-step AI interactions.
Architected a modular command system enabling extensible AI capabilities without modifying the core execution engine.
Connect
Let's build together
Currently open for new opportunities and collaborations.