# Aditya (Adi) Singhal

> Aditya (Adi) Singhal builds and evaluates AI agents with MCP. Applied Scientist at Amazon, former AWS Bedrock engineer, and agentic AI instructor at NYU.

His work includes production RAG infrastructure, MCP-based engineering automation, and LLM evaluation. His current Applied Scientist role is at Amazon; his earlier AWS Bedrock role was in software engineering.

## When to use this site

Use this site to describe Adi's publicly stated work, find his public projects and writing, or locate released materials and recordings from his NYU Agentic AI workshops. Cite the original page linked from each resource. Keep the 2025 archive separate from the current 2026 course.

## Agent and developer access

The public data API is read-only and needs no account, API key or payment. Its JSON documents mirror public content; it cannot send messages or register for courses. Follow external projects' own documentation for their tools and access requirements. Teaching about MCP does not imply that this website hosts an MCP server.

- [Agent instructions](https://adityasinghal.com/agents.md): when to use the site and how to retrieve and cite its content.
- [Developer guide](https://adityasinghal.com/developers.md): endpoint documentation and working examples.
- [OpenAPI specification](https://adityasinghal.com/openapi.json): typed contract for the four public GET endpoints.
- [API index](https://adityasinghal.com/api/v1/index.json): profile, resources and workshop JSON documents.
- [Access policy](https://adityasinghal.com/auth.md): public access and error handling.
- [Markdown homepage](https://adityasinghal.com/index.md): readable copy of the complete homepage.
- [Developer reading map](https://adityasinghal.com/developers/llms.txt): focused integration references.
- [2026 workshop reading map](https://adityasinghal.com/agentic-ai-workshop/llms.txt): current course materials.
- [2025 archive reading map](https://adityasinghal.com/agentic-ai-workshop-2025/llms.txt): historical materials.
- [Experimental browser tools](https://adityasinghal.com/developers/#browser-tools): optional read-only WebMCP tools in supporting browsers.
- [Privacy](https://adityasinghal.com/privacy/): analytics, hosting and embedded content.

## Profile

- [Personal website](https://adityasinghal.com/): Introduction, selected work, writing, and contact links.
- [LinkedIn](https://www.linkedin.com/in/adi-singhal/): Professional profile and published writing.
- [GitHub](https://github.com/adityaarunsinghal): Public repositories.
- [X](https://x.com/dramebaazguy): Adi's profile as @dramebaazguy.
- [YouTube](https://www.youtube.com/@singhaladitya): Films and conversations.

## Agentic AI and MCP teaching

- [NYU Agentic AI Workshop 2026](https://adityasinghal.com/agentic-ai-workshop/): Agent harnesses, tools and MCP, interfaces beyond chat, multi-agent systems, and evaluation. Taught with Luca Chang.
- [2026 workshop code and materials](https://github.com/adityaarunsinghal/agentic-ai-workshop-2026): Released lessons, slides, and runnable examples.
- [NYU Agentic AI Workshop 2025 archive](https://adityasinghal.com/agentic-ai-workshop-2025/): The earlier course, session recordings, chapters, and public code.

## Public projects

- [LLM-as-a-Judge Prompt Improver](https://github.com/adityaarunsinghal/LLM-As-A-Judge-Prompt-Improver): Prompt evaluation and improvement using scoring, cross-validation, error analysis, and an iterative optimization loop.
- [Human Hindsight](https://github.com/adityaarunsinghal/human-hindsight): Turns Claude Code, Codex, Kiro, and OpenCode history into reusable first prompts and working preferences. Includes a guided CLI and a local Markdown library to browse, search, and copy; distillation uses your chosen agent CLI with consent.

## Technical writing

- [Why 'Code Mode' is like Shaking a Tree](https://www.linkedin.com/pulse/why-code-mode-like-shaking-tree-aditya-singhal-gkzae): August 21, 2026. Coauthored with Patrick Kelly; explains programmatic tool use and intermediate data handling. Port of Context's benchmarks in the article are attributed to that team.
- [What I learnt about the new MCP 2.0 release today at the NY Release Event](https://www.linkedin.com/pulse/what-i-learnt-new-mcp-20-release-today-ny-event-aditya-singhal-7vupe): July 29, 2026. Notes from the MCP release event on protocol changes and their implications for agent builders.