The AI work started before t0rque—inside an R&D environment where prototypes had to become useful tools.

A model response was never the whole product. The work lived in the less theatrical parts: application architecture, prompt and agent behaviour, review boundaries, collaboration and the ordinary engineering required to make an experiment usable.

Research project

Foundry

The problem

Non-technical users could describe an application conversationally, but useful generation required more than returning a code-shaped answer.

The product decision

Treat the interaction as an agent-assisted product workflow that produces self-hosted applications, with product structure and engineering constraints carried through the conversation.

Akshay’s contribution

Proposed and implemented the platform inside Thomson Reuters Labs.
Owned React application architecture while contributing Python FastAPI services and data structures.
Worked with data scientists, UX designers and engineers to make an ambitious prototype usable and integrable.

What happened

The initiative was adopted and integrated into broader enterprise tools. The available résumé does not establish usage scale or commercial impact, so neither is inferred here.

Responsible boundary

The public case explains Akshay's product and engineering contribution without exposing internal architecture, prompts or proprietary workflows.

Research project

ContentCraft

The problem

Document-processing automation could accelerate content operations, but generated content still needed an accountable quality and compliance boundary.

The product decision

Keep people inside the workflow: automate document work and make review of LLM-generated content an explicit product stage rather than an informal cleanup task.

Akshay’s contribution

Spearheaded the agentic workflow and contributed across frontend, backend and prompt-system behavior.
Designed the experience around document processing, generated output and human review.
Collaborated across engineering, data science and design while maintaining code and interaction quality.

What happened

The workflow gave the content operations team a structured way to automate processing while retaining human review. No unsupported adoption or performance metric is added.

Responsible boundary

Human review remained responsible for quality and compliance decisions; the model accelerated work without becoming the final authority.

What the environment taught

R&D is not permission to ignore the product.

01Prototype the uncertain capability.
Use working software to discover what the argument or diagram missed.
02Keep the whole application in view.
Agent behaviour, frontend interaction, APIs, data and review rules shape the same experience.
03Work across disciplines.
Useful AI products require engineering, data science, design and domain judgment to disagree productively.
04Give judgment a visible boundary.
Human review is part of the system when quality or compliance changes the consequence.

Public case-study boundary

This account uses the project names and outcomes documented in Akshay’s résumé. It does not expose proprietary prompts, internal architecture or usage metrics that were not supplied.