Akshay

Contact

AI systems / developer tools / cloud workflows

I build AI systems and developer tooling that make messy workflows usable.

I am Akshay, an engineer in Boston working across agentic AI, developer tooling, workflow automation, and cloud development environments. This site is the public-safe layer behind the work: selected repositories, anonymized systems notes, and the engineering tradeoffs worth writing down.

Selected Work

Compact case studies from public repos and public-safe systems work. Each item focuses on what the system does, where the interesting boundary is, and what it proves technically.

Developer tools

Raycast Vim Leader Key

A Raycast extension for Vim-style shortcut sequences: nested groups, single-key navigation, quick capture, browser-aware URL routing, and import/export for shared configs.

  • Improves repeated desktop workflows without requiring global shell scripts.
  • Published on the Raycast Store.
  • Includes migration and conflict handling for hierarchical shortcut configs.
  • Current local work strengthens active-browser detection and test coverage.
Raycast TypeScript Node tests
AI systems

Anonymized AI travel platform

A conversational booking system combining chat, voice, travel search, payments, and observability into a full-stack AI product slice.

  • Agent graph connects planning prompts with Amadeus flight and hotel tools.
  • Voice sessions use STT/TTS integration with WebSocket context updates.
  • Telemetry covers request latency, provider errors, tool calls, and token usage.
FastAPI Next.js LangGraph Stripe
Document AI

Document Intelligence Platform

A cloud-native Python platform for document upload, extraction, semantic search, summarization, async processing, and API-driven document workflows.

  • Separates API, services, repositories, clients, and background jobs.
  • Pairs FastAPI and Celery with object storage, PostgreSQL, Redis, and vectors.
  • Documents the production shape: Docker, Kubernetes, typing, linting, tests.
Python FastAPI Celery Vector search
Cloud environments

Modal, Hermes, and DevPod tooling

Local experiments for running agent tools in remote containers, preserving runtime state, and standardizing reusable cloud development environments.

  • Compares Modal functions with runtime sandboxes for isolated code execution.
  • Runs Hermes in Modal with persistent volume-backed config and memory state.
  • Defines DevPod templates for Python and polyglot full-stack environments.
Modal DevPod Docker Python

Work Map

The portfolio is organized around systems I keep returning to: AI workflows, developer tools, and reproducible cloud environments.

AI systems

Agent graphs, voice/chat surfaces, document intelligence, semantic search, and observability for model-backed workflows.

See case studies

Developer tools

Raycast extensions, browser automation, and small utilities that turn repeated work into explicit interfaces.

View GitHub

Cloud/dev environments

Container-first development templates, remote execution sandboxes, and agent runtime experiments that make environments reproducible.

Read notes

Notes

Planned technical write-ups. These are the topics I would rather explain with constraints, rejected options, and failure modes than with a list of buzzwords.

Planned

Shortcut sequences as interface design

What a Raycast leader-key workflow teaches about hierarchy, capture friction, migration, and browser-aware desktop automation.

Planned

Voice agents need boring observability

What to measure when a conversational product crosses model calls, provider tools, WebSockets, STT/TTS, and payments.

Planned

Remote sandboxes as agent infrastructure

Notes from comparing predeclared remote functions with runtime-created containers for code execution and isolated agent work.