Skip to content
Aelsha Innovations

Company

Aelsha Innovations: an AI engineering and applied research company.

We work with organisations putting AI agents to work. We design, build, secure and observe agents inside the software they already run, and we research the questions that work raises: how to evaluate an agent, how to secure it, and how to keep it answerable to the people it works for. Every engagement is delivered with AELSHA™ ASTRA, our framework for taking agents to production and keeping them there. When we publish an evaluation of our own, it carries the method, the results and the limits.

Our mission: To build technology that acts for the people we serve, answers to them, and is judged by their success.

Today that means AI agents, connected to the software an organisation already runs, and its own people deciding where those agents may act.

The four working rules around one agent system. The person-approves rule is the single gold node.

What the diagram shows ↓

Intelligence you can use. Autonomy you can prove.

AI agents now act inside the systems a business already runs. In what we build, the proof is yours to check: every step traced, behaviour tested on your own examples before each release, a person approving each consequential action, and an off switch.

How we work

Four rules that hold for every agent system we build.

  1. Tested on the client’s own cases

    Done is defined before design begins: the client’s examples, checked before each release and left as tests the client’s own team can keep running.

  2. Only the access the job needs

    Each agent and tool gets the access and data its task requires, and nothing more.

  3. A person approves what matters

    Agents propose and prepare; a person decides before anything consequential changes, and can stop it.

  4. Every step on the record

    Each action an agent takes is traced, so what happened can be explained, and improved, by the client’s own team.

How we handle security and data →

Our vision

A world where the more machines do, the more clearly humans see that work and the more surely they steer it.

Our values

Four values, each something you could watch us do.

Their outcome, not our output
We measure an engagement by what changed for the client, and for the people their system touches, not by what we shipped. When a smaller system does the job, we propose the smaller system.
Read it first-hand
We read the paper, the code and the logs before we form a view. What we don’t yet know, we research rather than repeat, and we show our working.
Say the hard thing early
We raise a problem when we find it, not at the end. We state what we don’t know and the limits of what we measured, and if the honest answer is not to build, we say so before money is spent.
Simple outlasts clever
Complexity is our job, not our client’s. We choose the plain design that can be explained and maintained years from now over the clever one that can’t.

How we research

Our research follows our engineering. On evaluation: whether a checker written in code, with no second model as judge, can catch invented names in AI-written documentation; that was our first published evaluation, of AELSHA™ NEHEMVIX. On security: how an agent is attacked through what it reads, set out as the ten-threat matrix on our agent-security page, mapped to the OWASP Top 10 for Agentic Applications, the MCP specification and MITRE ATLAS. On governance: what an agent may do alone before a person decides, the autonomy tiers in AELSHA™ ASTRA. Findings appear in Insights under the label Research, each stating its question, method, data, results and limits. They are our own work, not independent audits.

Read the research in Insights →

Why we do this work

Machines have begun to do work that once needed a person: reading, deciding and changing things inside the systems organisations run on. The pressure is to build it fast. We think it is worth building only if the organisation it works for can check what it did, correct it, and remain the party that decides, and that building it that way is harder, and worth more, than building it fast. We research and build so that organisations take this step on evidence of their own, not on our say-so.

Leadership

Who answers for this work.

Gladys Esmeralda

Founder & Director

The legal entity

The company behind this website, AELSHA™ ASTRA and AELSHA™ NEHEMVIX is a private limited company registered in India.

Registered name
AELSHA INNOVATIONS (OPC) PRIVATE LIMITED
Corporate Identity Number (CIN)
U62099TN2024OPC175379
Incorporated
13 December 2024
Registered office
No. 25, Second Street, Bakthavatchalapuram, Avadi, Chennai 600054, Tamil Nadu, India.
Contact for queries and grievances
Gladys Esmeralda, Director
Email for queries
support@aelsha.com
Email for grievances
security@aelsha.com

Grievances and requests about your personal data go to security@aelsha.com, which our Grievance Officer reads; for anything else, write to support@aelsha.com.

Aelsha Innovations at a glance

Aelsha Innovations · AI engineering & applied research

Aelsha Innovations is an AI engineering and applied research company, with agentic AI as its flagship specialisation. It designs, builds, secures and observes production AI agent systems, and researches how to evaluate, secure and govern them; each published evaluation carries its method and its limits. It delivers every engagement with AELSHA™ ASTRA, its framework for taking AI agents to production and keeping them there: lifecycle gates, tracing, evaluation and governance. AELSHA™ NEHEMVIX, a free tool, explains unfamiliar codebases; you bring your own AI model key.

Updated

Tell us what you want to build or secure

Describe the problem in a few lines. A person reads it and replies by email.

Talk to us about your project