New opportunity

AI Application Developer & Automation Engineer

Astrobase Space Technologies · Bengaluru, India

About this role

Astrobase Space Technologies seeks a hands-on developer in Bengaluru to build AI-powered applications and intelligent automations for its internal operations. The role focuses on agentic workflows and a custom AI-first web application with analytics, security and access controls, while maintaining existing Microsoft Power Platform solutions.

Responsibilities

Build agentic automations for document processing, data extraction, database analysis and cross-functional reporting. Integrate LLMs with internal systems through APIs, RAG pipelines and vector databases.
Architect, develop, test and progressively release the web application module by module. Design database schemas, implement row-level security and role-based access control, integrate authentication with Microsoft Entra ID, and build in-app analytics and reporting.
Fix and enhance Power Apps and Power Automate solutions, migrating workflows to the new platform where appropriate. Document code, prompts, workflows and architecture.

Requirements

The posting lists 1–3 years of experience and specifies up to 3 years in its qualifications. Hands-on AI application and agentic automation experience is required, with a portfolio or GitHub profile.
Proficiency with LangChain, LangGraph, CrewAI, AutoGen or an equivalent agentic framework; working knowledge of LLM APIs, prompt engineering, RAG, embeddings and vector databases such as Pinecone, Weaviate, Chroma or pgvector.
Strong full-stack development skills using React/Next.js or equivalent, Node.js or Python backends such as FastAPI/Flask, and REST APIs. Understand relational and NoSQL databases, schema design, query optimization, access controls and identity management. Familiarity with Power Apps, Power Automate, Dataverse, Git, testing and CI/CD is expected.

Preferred

Evidence of shipping a real product or project; exposure to ERP, PLM, MES, CRM or B2B SaaS; cloud experience, preferably Azure though AWS or GCP is acceptable; and MLOps basics. This is a solo-builder role reporting directly to leadership, requiring comfort with ambiguity and ownership.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Agentic AILangChainLangGraphCrewAIAutoGenLLM APIsOpenAIAnthropicRAGPrompt engineeringEmbeddingsPineconeWeaviateChromapgvectorReactNext.jsNode.jsPythonFastAPIFlaskREST APIsRelational databasesNoSQL databasesDatabase schema designQuery optimizationRow-Level SecurityRole-Based Access ControlMicrosoft Entra IDPower Apps by Microsoft Power Platform - developed apps with low code platform

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Build useful agentic automations
The role calls for workflows for document processing, data extraction, database analysis and cross-functional reporting, with LLMs connected to internal systems through APIs, RAG and vector databases. Prepare to explain a hands-on implementation and its results; the posting requires a portfolio or GitHub profile.
Ship the AI-first application incrementally
Astrobase expects the developer to architect, build, test and deploy a custom web application module by module, including in-app analytics and reporting. Prepare to explain how you would define a first module and verify it before expanding.
Make data access part of the design
Database schemas, row-level security, role-based access control and M365 Entra ID authentication are explicit responsibilities. Prepare to explain where permissions would be enforced across the application, API and data layer.
Evolve existing Power Platform workflows
The role includes fixing and enhancing Power Apps and Power Automate solutions, then gradually migrating workflows where it makes sense. Prepare a framework for deciding what to retain, repair or move.
Own maintainable delivery
The posting describes a solo-builder role reporting to leadership and requires documentation of code, prompts, workflows and architecture. Prepare to show how you organize testing, releases and documentation while working through ambiguous requests.

Likely challenges to think through

Automation versus trustworthy output
Possible challenge: LLM-assisted extraction and analysis can save manual effort, but incorrect or incomplete outputs could undermine internal reporting. Prepare to discuss validation, failure handling and how a workflow would signal uncertainty rather than silently pass along a result.
Speed versus secure phased releases
Possible challenge: releasing modules progressively may create pressure to defer schema and access-control decisions that are harder to change later. Prepare to describe a minimal secure foundation and the tests you would run before each release.
New platform versus existing workflows
Possible challenge: building the new application while supporting Power Platform solutions could split limited development time. Prepare to explain how you would assess workflow value, migration effort and continuity before choosing what to move.
Ownership versus uncertain scope
Possible challenge: a solo-builder reporting to leadership may need to turn broad operational requests into a feasible sequence of work. Prepare to describe how you would clarify users, data access and success measures before committing to a module.

Practice your answers

Technical practice

How would you design an agentic workflow for document processing and data extraction?

What to coverPreparation suggestion: Explain how you would divide the work into steps, validate extracted fields, and handle uncertain or failed outputs. Discuss when a framework such as LangGraph or an equivalent helps more than a simpler workflow.

How would you use RAG to answer questions from internal documents?

What to coverPreparation suggestion: Cover document preparation, embeddings, retrieval from a vector database, and grounding answers in retrieved content. Explain how you would check retrieval quality and respond when the evidence is insufficient.

How would you integrate an LLM with an internal system through an API?

What to coverPreparation suggestion: Describe the API boundary, prompt and response handling, and controls on what data or actions the model can access. Address failures such as malformed outputs, timeouts, and unintended actions.

How would you structure a full-stack module for the proposed AI-first web application?

What to coverPreparation suggestion: Describe responsibilities across a React/Next.js or equivalent frontend, a Node.js or Python backend, and REST APIs. Explain how you would keep the module testable and suitable for progressive release.

How would you design and tune a database for internal application workflows?

What to coverPreparation suggestion: Explain how the data and query patterns would guide schema design and a relational-versus-NoSQL choice. Discuss how you would identify and improve a slow reporting or application query.

How would you combine Entra ID authentication with RBAC and row-level security?

What to coverPreparation suggestion: Distinguish identity verification, permitted actions, and permitted records. Explain where you would enforce each control and how you would test that one role cannot access another role's data.

How would you build trustworthy in-app analytics and cross-functional reporting?

What to coverPreparation suggestion: Describe how you would define metrics, derive them from application data, and check that reports remain consistent with source records. Address access controls so a useful aggregate report does not expose restricted data.

How would you test and progressively deploy modules of the new application?

What to coverPreparation suggestion: Outline tests for APIs, permissions, and AI-assisted workflows, including cases where model output varies. Explain how Git and basic CI/CD checks could support a controlled module-by-module release.

When would you fix a Power Automate workflow rather than migrate it?

What to coverPreparation suggestion: Compare the immediate fault, dependencies, maintenance burden, and value of moving it to the new platform. Explain how you would verify that a fix or migration preserves the workflow's intended result.

What would you document for an agentic workflow so it remains maintainable?

What to coverPreparation suggestion: Cover the workflow's steps, prompts, API and data dependencies, failure handling, and relevant architecture decisions. Explain how you would keep that documentation aligned with changes to the implementation.

Behavioural practice

Tell me about a time you made progress on an ambiguous development task.

What to coverPreparation suggestion: Use a real situation, the questions you resolved, and the actions you took without waiting for every detail to be specified. End with the result and what you communicated to stakeholders.

Tell me about something you shipped beyond a prototype or coursework.

What to coverPreparation suggestion: Describe the need, your specific contribution, and how you tested and released it. Give an observable result rather than claiming impact you cannot substantiate.

Tell me about a time you aligned people from different functions on a reporting or automation need.

What to coverPreparation suggestion: Set out the differing needs, how you clarified definitions or priorities, and what you implemented. Close with the result and any feedback or correction that followed.

Tell me about a time you diagnosed and fixed an existing workflow or application.

What to coverPreparation suggestion: Explain the symptoms, how you isolated the cause, and why you chose a fix. State how you checked that it worked without disrupting other users or processes.

Tell me about a time you balanced a new build against maintenance work.

What to coverPreparation suggestion: Describe the competing demands, how you assessed urgency and impact, and how you communicated your decision. Give the result, including any trade-off or follow-up work.

Sharpen your knowledge

Topics to prepare

Agentic workflow design
The posting calls for agentic automations for document processing, data extraction, database analysis and reporting. As revision, sketch one workflow’s steps, tool calls, failure paths and points for human review.
LLM integration and retrieval
LLM APIs, prompts, RAG, embeddings and vector databases are explicit requirements. As revision, trace how a document becomes searchable and how retrieved passages reach an LLM; identify where an inaccurate answer could arise.
Full-stack application architecture
The role includes a custom AI-first web application released module by module, with REST APIs and in-app analytics. As revision, outline a first module’s frontend, backend, API, data flow and reporting view.
Database design and query performance
The posting names relational and NoSQL databases, schema design and query optimization. As revision, compare two data models for internal records and explain how you would investigate a slow reporting query.
Identity and data access
Row-Level Security, Role-Based Access Control and M365 Entra ID authentication are stated responsibilities. As revision, distinguish authentication, role permissions and row-level filtering using a two-role access matrix.
Power Platform evolution
The role supports Power Apps, Power Automate and Dataverse, with gradual migration where appropriate. As revision, compare fixing an existing workflow with moving it to the new application, including continuity and testing considerations.

Skills to revise

Build and test an agentic automation
As a practice exercise, use one of the listed agentic frameworks or an equivalent to process sample documents into structured data and a short report. Test missing fields and failed calls; this reflects the posting’s automation work.
Prototype a grounded LLM response
As a practice exercise, connect an LLM API to a small RAG pipeline over sample documents, then check whether answers reflect retrieved material. This exercises the stated LLM, prompt, embedding and vector-database knowledge.
Deliver a small web module
As a practice exercise, build a frontend and REST-backed module with a simple analytics view, then test it as an independently releasable increment. The posting calls for full-stack development, reporting and progressive deployment.
Implement and verify access rules
As a practice exercise, model records for two roles, implement RBAC and row-level restrictions, and test that each role sees only permitted rows. Map where M365 Entra ID authentication would supply identity; its integration is specified in the posting.
Maintain a Power Platform workflow
As a practice exercise, review a sample Power Apps and Power Automate flow, propose a bug fix, and document whether its workflow should remain or migrate. This matches the stated support and gradual-migration responsibility.
Show maintainable delivery
As a practice exercise, put a small module in Git with tests and concise documentation for its code, prompts, workflow and architecture; outline a basic CI/CD check. Git, testing, CI/CD basics and documentation are named in the posting.

Your preparation checklist

Select a portfolio walkthrough
The posting requires a portfolio or GitHub profile. Choose an AI application or automation you built and prepare a brief walkthrough of your contribution, architecture, tests, deployment status and observed results. Distinguish shipped work from prototypes.
Explain an agentic workflow
Prepare a diagram for document processing, data extraction or reporting, all tasks named in the posting. Show where an agentic framework, LLM API and internal-system API would fit, and how you would handle errors and review outputs. Label any unbuilt design as hypothetical.
Trace a RAG request end to end
Be ready to explain how documents become embeddings, enter a vector database and are retrieved for an LLM response. Prepare a small example showing how you would test retrieval quality and avoid exposing information a user cannot access.
Sketch a secure application module
As a design exercise, outline one module of the advertised AI-first web application: frontend, backend API, schema, analytics view and tests. Show how M365 Entra ID authentication, role-based access control and row-level security would work together.
Review Power Platform maintenance decisions
Prepare an example of diagnosing a Power Apps or Power Automate issue, or a hypothetical troubleshooting plan if you lack a relevant example. Explain what evidence would justify enhancing an existing workflow versus migrating it to the new platform.
Show how you deliver and document changes
Prepare evidence from a real project, where available, of Git use, testing, CI/CD and documentation. Then outline how you would release the web application module by module and record code, prompts, workflows and architecture for maintainability.
Write questions for the employer
Ask which internal workflow is the first priority, what systems and data it must connect to, how access permissions are managed today, and who will review module releases. These questions clarify the scope of the posting’s solo-builder role without assuming a particular implementation.

Availability last checked: 5 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…