New opportunity

Data Scientist

UptimeAI · Bangalore, India

About this role

UptimeAI is seeking a Data Scientist to bridge algorithmic research and product engineering for industrial AI reasoning agents. Based in Bangalore, Karnataka, this is a full-time hybrid role focused on turning complex operational data into intuitive, actionable AI experiences and production-grade systems.

Responsibilities

Convert early-stage machine-learning models and RAG pipelines into interactive, full-stack alpha products within days; build and deploy AI systems beyond prototypes.
Design agentic workflows and front-end interfaces through which plant engineers can interact with root-cause-analysis agents and explore time-series plots.
Analyze noisy, incomplete and real-time customer data from SCADA, DCS and IoT systems, then present findings in functional user interfaces.
Work with data scientists to understand model constraints, engineers to make prototypes scalable through API-first and containerized designs, and industrial domain experts to create context-aware solutions.
Apply design thinking to operational workflows such as maintenance scheduling and process stability; solve ambiguous problems and translate customer needs into working products.

Qualifications and skills:

The posting specifies 3–5 years of experience and requires 3+ years in a role combining data science and software engineering, such as ML engineering, product engineering or data-science prototyping.
Bachelor’s degree or higher in Chemical Engineering, Mechanical Engineering, Aerospace or Applied Physics, emphasizing computational modeling. Candidates with computer science degrees are welcome if they have significant heavy-industry experience.
A portfolio of end-to-end AI applications used through an interface; strong Python, Polars or Pandas, and Scikit-learn skills; ability to adjust ML models or optimize RAG prompts.
Proficiency with Cursor, Antigravity or similar AI-integrated development tools; ability to manage data flow from PostgreSQL or TimescaleDB through FastAPI to a polished frontend. Industrial judgment, customer-facing product sense and rapid problem-solving are important.

Preferred

Prior experience in industrial AI, manufacturing or energy is highly preferred. The role offers opportunities to influence product development and lead industrial AI projects.

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

Skills for this role

PythonPolarsPandasScikit-learnMachine LearningRAGAI AgentsCursorAntigravityClaude CodePostgreSQLTimescaleDBFastAPIExploratory Data AnalysisSCADADCSIoTAPI DesignContainerizationFrontend DevelopmentUX DesignDesign ThinkingComputational ModelingTime Series AnalysisRoot Cause Analysis

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

Turn research into usable products quickly
The role calls for converting early-stage ML models and RAG pipelines into interactive, full-stack alpha products within days, while also building deployable AI systems. For preparation, select an end-to-end application you can walk through from model or pipeline to an interface people used.
Translate industrial data into visible insights
The posting emphasizes EDA on noisy, incomplete, real-time SCADA, DCS and IoT data, with findings reflected in a functional UI. Prepare to explain how you distinguished meaningful operational signals from data-quality issues and presented the result.
Make agentic reasoning usable for plant engineers
A stated responsibility is designing interfaces for interactions with root-cause-analysis agents and time-series plots. Prepare to discuss how an engineer would investigate an agent’s finding and understand the evidence behind it.
Build a path from alpha to scalable system
The role bridges Data Science model constraints and Engineering’s need for API-first, containerized designs. The posting also describes data flow from PostgreSQL or TimescaleDB through FastAPI to a frontend. Prepare to explain architectural choices, not just a demonstration.
Shape solutions around operational workflows
The posting calls for working with industrial domain experts and customers to solve ambiguous needs, including maintenance scheduling and process stability. Prepare an example of how domain input changed the problem definition or the product design.

Likely challenges to think through

Speed versus production readiness
Likely preparation hypothesis, not a confirmed employer problem: delivering an alpha within days can compete with the expectation of deployable, scalable systems. Prepare to explain what you would validate first, what you would defer, and how you would keep an initial build extensible.
Sensor drift versus consequential change
Likely preparation hypothesis: incomplete, real-time data can make a small sensor change either noise or an important operational signal, as the posting illustrates. Prepare to describe checks against context and data quality, and how a UI could communicate uncertainty without hiding the observation.
Agent flexibility versus understandable decisions
Likely preparation hypothesis: an agent that reasons, adapts and suggests actions may be difficult for a plant engineer to assess during root-cause analysis. Prepare to explain how an interaction would expose relevant time-series evidence and model limitations while leaving room for domain-expert judgment.

Practice your answers

Technical practice

How would you explore a noisy, incomplete SCADA, DCS or IoT dataset before building a user-facing feature?

What to coverSuggested preparation: explain how you would check timestamps, missing values, sensor quality and operating context using Python and Polars or Pandas. Describe which findings belong in a functional UI and which uncertainties should remain visible.

How would you decide whether a small sensor drift indicates a process issue or ordinary noise?

What to coverSuggested preparation: discuss baselines, operating conditions, related signals and engineering context rather than relying on a fixed percentage threshold. Explain how you would communicate the evidence and uncertainty to a plant engineer.

How would you adjust and evaluate an early-stage machine-learning model for industrial time-series data?

What to coverSuggested preparation: describe a plausible Scikit-learn workflow, including features, a time-aware evaluation split and checks for data leakage. Discuss the trade-off between useful detection and false alarms in an operational setting.

How would you improve a RAG pipeline whose answers are not useful for root-cause analysis?

What to coverSuggested preparation: distinguish retrieval problems from prompt or model-response problems, then explain what you would change and how you would compare results. Address whether answers are grounded in available evidence and clear about uncertainty.

How would you design an agentic root-cause-analysis workflow for a changing industrial situation?

What to coverSuggested preparation: outline how the agent would incorporate new observations, present competing explanations and decide when to seek engineer input. Discuss the trade-off between useful proposed actions and keeping consequential decisions understandable to the user.

What should a plant engineer be able to do when an RCA agent is paired with a time-series plot?

What to coverSuggested preparation: describe interactions that let the engineer inspect time windows and signals while questioning the agent's explanation. Explain how the interface would connect a proposed cause to plotted evidence without overstating confidence.

How would you move industrial time-series data from PostgreSQL or TimescaleDB through FastAPI to a frontend?

What to coverSuggested preparation: outline the data flow, API contract and handling of queries over selected time ranges. Explain trade-offs between responsive exploration, data freshness and the volume of data sent to the interface.

How would you turn an early ML or RAG prototype into an interactive alpha within days without blocking later deployment?

What to coverSuggested preparation: identify the smallest end-to-end user workflow and what you would defer. Explain how API-first boundaries, containerization and checks around model constraints could make the prototype easier to extend.

How would you use an AI-integrated development tool to build a full-stack feature quickly while checking its output?

What to coverSuggested preparation: explain which boilerplate or integration tasks you might accelerate with a tool such as Cursor, Antigravity or Claude Code. Describe how you would review generated code and test the data flow and user-facing behavior.

How would you determine whether to build for maintenance scheduling or process stability when a customer's need is still vague?

What to coverSuggested preparation: use the posting's design-thinking focus to explain how you would examine the operational workflow and available data before choosing a feature. Describe a small interface or analysis you could use to test whether the proposed solution addresses the customer's problem.

Behavioural practice

Tell me about a time you turned an ambiguous customer need and messy data into a working product direction.

What to coverSuggested preparation: use a real situation, your actions to clarify the operational problem, and the resulting decision or product outcome. Distinguish what the data established from what still needed customer validation.

Describe a time a domain expert changed your interpretation of an analytical result.

What to coverSuggested preparation: explain the situation, how you incorporated the expert's reasoning, and what changed in the analysis or interface. Close with the result and any remaining uncertainty.

Tell me about a time you reconciled a model constraint with engineering needs for a deployable feature.

What to coverSuggested preparation: describe the situation, your role in the discussion with data science and engineering colleagues, and the trade-off you made. State the result without implying that speed, accuracy and scalability were all maximized.

Describe a time you delivered an interactive prototype under a tight deadline.

What to coverSuggested preparation: give the situation, the scope you chose, the actions that made an end-to-end demonstration possible, and the result. Explain what you deliberately left for later and why.

Tell me about a time user feedback changed how you presented a complex technical finding.

What to coverSuggested preparation: identify the user's workflow, the feedback you received, and the specific interface or explanation you changed. Describe the result using observable evidence rather than assuming the change improved decisions.

Sharpen your knowledge

Topics to prepare

Industrial sensor interpretation
The posting emphasizes noisy, incomplete, real-time SCADA, DCS and IoT data, plus judgment about whether sensor drift is significant. Revision exercise: reason through a hypothetical 2% sensor drift, listing what operating context and data-quality checks you would need before calling it a fault.
Agentic root-cause analysis and RAG
The role calls for agents that reason, adapt and take actions, and for turning RAG pipelines into interactive products. Revision exercise: diagram how an RCA agent would use retrieved context, handle uncertain evidence and present a proposed explanation for an engineer to inspect.
Model constraints and computational reasoning
The posting asks for a strong ML foundation and collaboration with data scientists on model constraints; it also emphasizes computational modeling. Revision exercise: compare two possible explanations for a process-stability change and note what measurements or model limitations would affect each conclusion.
Prototype-to-production architecture
The job distinguishes rapid full-stack alpha products from production-grade systems and specifies API-first, containerized designs. Revision exercise: sketch the components and data flow of an alpha product, then identify what you would change before deployment.
Design thinking for plant workflows
The posting names maintenance scheduling, process stability and collaboration with plant or reliability experts. Revision exercise: turn a hypothetical engineer's ambiguous concern into a user question, an actionable screen and the domain questions you would ask before building it.

Skills to revise

Explore messy time-series data
The role explicitly calls for EDA on customer datasets and names Python with Polars or Pandas. Practice loading a sample time-series dataset, checking missing values and timing irregularities, and summarizing findings that should appear in a UI.
Adjust an ML model or RAG prompt
The posting names Scikit-learn and the ability to tweak a model or optimize a RAG prompt. Practice making one controlled change, comparing outputs on the same examples, and explaining its effect and limitations.
Connect database, API and frontend
The specified data path is PostgreSQL or TimescaleDB through FastAPI to a polished frontend. Practice tracing or building a small flow that retrieves time-series records, exposes them through an API and displays them; explain how errors reach the user.
Design an engineer-facing RCA interaction
The role asks for interfaces where a plant engineer can talk to an RCA agent and probe a time-series plot. Sketch a screen that links an agent's explanation to the relevant plot interval and makes uncertainty visible.
Build rapidly with an AI-integrated IDE
Cursor, Antigravity or similar AI-integrated tools are explicitly named for fast, integrated development. Practice using one to assemble a small feature, then review generated code, test its data flow and outline how you would containerize it; speed does not replace verification.
Present an end-to-end application
The posting requests a portfolio of AI applications people have used through an interface. If you have a relevant example, rehearse a concise walkthrough from user problem and messy data to model constraints, interface decisions and deployment; do not imply usage you cannot substantiate.

Your preparation checklist

Select an end-to-end application walkthrough
The posting asks for AI applications people have used through an interface. If available, prepare a demo or screenshots showing the user workflow, your contributions, model or RAG component, and what reached deployment; label any illustrative work as such.
Practice an industrial-data analysis
Using a suitable sample dataset, prepare a short Python and Polars or Pandas walkthrough of missing values, noisy sensors, time-series drift, and an insight you would show in a UI. SCADA, DCS and IoT data are named in the role; do not imply the sample is customer data.
Sketch a root-cause-analysis interaction
As a preparation exercise, draw a screen where a plant engineer questions an RCA agent and probes a time-series plot. Annotate how the interface distinguishes a potentially important sensor drift from noise and communicates uncertainty.
Explain a model or RAG improvement
Prepare one concrete example, if available, of adjusting a Scikit-learn model or RAG prompt: the constraint, change, evaluation and trade-off. The posting expects comfort with both ML adjustments and RAG prompting, not a specific algorithm.
Diagram the path from alpha to deployable product
Show how an early ML or RAG pipeline could move through PostgreSQL or TimescaleDB, FastAPI and a frontend. Mark API boundaries, containerization and the checks you would make with data science and engineering before scaling; these are design choices to discuss, not a claimed company architecture.
Organize evidence for the stated background
Prepare concise examples of work blending data science and software engineering, computational modeling or relevant heavy-industry experience, and rapid development with an AI-integrated IDE such as Cursor or Antigravity. Distinguish direct experience from areas you would need to learn.
Bring questions about the operating workflow
Ask which maintenance-scheduling or process-stability decisions users need help making, what constraints real-time plant data presents, and how domain experts, data science and engineering decide an alpha is ready to scale.

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…