Sentiaflow

About Sentiaflow

Sentiaflow is an AI engineering and IT services company building production-grade agentic AI systems, sitting at the intersection of LLMs and real business operations — data, APIs, permissions, workflow state, human decisions, security, and measurable outcomes. Our initial domain focus is healthcare, particularly clinical trials, where reliability, traceability, and clear human-decision boundaries matter more than a slick demo.

Job Description

As a Level 1 engineer, you'll implement bounded parts of a production agentic workflow under an Agent Captain or senior engineer, connecting models to application services, tools, data sources, and human approval points.

You will:

• Translate a scoped business workflow into typed inputs, outputs, states, actions, and escalation paths
• Build backend services and tool integrations (Node.js/TypeScript or Python)
• Use LLMs only where model judgment adds value; keep rules, validation, authorization, and workflow control in deterministic code
• Design and validate structured model outputs before they touch downstream systems
• Handle partial data, tool failures, duplicate events, retries, timeouts, rate limits
• Add logging, traces, metrics, and decision records for diagnosability
• Write tests and evaluation cases that check whether the full workflow behaves correctly — not just whether output sounds fluent
• Protect sensitive data; participate in code, design, and release-readiness reviews
• Explain implementation trade-offs clearly to engineers and stakeholders

Success in 6 months: own a bounded workflow module end-to-end, integrate models without letting probabilistic output bypass deterministic controls, produce release-ready evaluation evidence, and diagnose cross-boundary failures with less supervision.

Desired Skills

• Approximately 3–6 years hands-on backend/application engineering experience, with demonstrable hands-on work building agentic systems — not just calling an LLM API from a backend service
• LangGraph (or comparable agent orchestration framework) experience is required — building multi-step, stateful agent workflows with conditional branching, tool-calling loops, and recovery/retry logic, not a single-prompt wrapper
• Deep RAG experience, including:
• Chunking strategy design, embedding model selection, and retrieval evaluation (not just "connected a vector DB")
• Hybrid search (dense + sparse/keyword), re-ranking, and query rewriting/decomposition
• Handling retrieval failure modes: irrelevant context, stale data, contradictory sources, citation/grounding accuracy
• Measuring RAG quality (precision/recall on retrieval, faithfulness/groundedness of generation) — not eyeballing outputs
• Experience designing agent state machines / workflow graphs: tool selection, planning loops, human-in-the-loop interrupts, checkpointing, and state persistence across long-running workflows
• Strong programming in JS/TypeScript (preferred), Python, Java, C#, or Go
• Solid grasp of API design, databases, async processing, auth, testing, deployment
• Comfort reasoning about state, retries, idempotency, concurrency, permissions, audit trails, failure recovery
• Real production debugging experience, not just greenfield builds
• Clear technical communication

We're looking for engineers who've actually built and tuned agentic/RAG systems in production — not those who've only wired together frameworks or prompted an LLM API.

Nice to have: experience with other orchestration frameworks (CrewAI, AutoGen, custom state machines), observability/eval tooling (LangSmith, Langfuse, custom trace pipelines), healthcare or regulated-industry background. Bachelors from IIT or NIT highly preferred.

Agent Application Engineer (Backend / Agentic AI) — Level 1

About Sentiaflow

Sentiaflow is an AI engineering and IT services company that designs and delivers production-grade agentic AI systems. We work at the point where large language models meet real business operations: data, APIs, permissions, workflow state, human decisions, security, evaluation, and measurable outcomes.

Our initial domain focus is healthcare, particularly clinical trials and related operational workflows. These are environments where an impressive demo is not enough. A system must be dependable, traceable, appropriately controlled, and clear about when a human must make the decision.

We are building a specialised engineering organisation across three disciplines:

• Agent Application Engineering
• Agent Platform & Evaluation Engineering
• Applied AI & Model Engineering

The Role

We are hiring strong backend or application engineers who want to specialise in building agentic AI applications. You do not need to arrive with several years of "agent engineer" experience. We are more interested in whether you have built real software, understand how production systems fail, and can learn to use models inside controlled business workflows.

As a Level 1 Agent Application Engineer, you will implement bounded parts of a production workflow under the guidance of an Agent Captain or senior engineer. Your work will connect models to application services, tools, data sources, and human approval points. You will be expected to make every important step observable, testable, and recoverable.

This is not a prompt-writing position, and it is not a generic chatbot role. It is an application engineering role for systems in which some decisions are probabilistic, while the surrounding controls must remain deterministic.

What You Will Work On

The specific client problem will vary, but typical work may include:

• Converting a clinical-operations signal into a bounded workflow that gathers evidence, analyses likely causes, presents options, and routes a recommendation to an authorised person.
• Building services that extract or structure information from trial documents and pass the result through validation and human review.
• Integrating agent workflows with internal APIs, databases, document systems, notification services, and client platforms.
• Implementing workflow state, recovery behaviour, approval gates, permissions, traceability, and scenario-based evaluations.

The work concerns operational decision support and workflow execution. It does not delegate clinical judgment or patient-care decisions to an autonomous model.

What You Will Be Responsible For

• Translate a clearly scoped business workflow into typed inputs, outputs, states, actions, and escalation paths.
• Build reliable application services and tool integrations using Node.js/TypeScript or Python.
• Use an LLM only where model judgment adds value; implement rules, validation, authorisation, and workflow control in deterministic code.
• Design structured model outputs and validate them before they can affect downstream systems.
• Handle partial data, unavailable tools, duplicate events, retries, time-outs, rate limits, and other normal production failure modes.
• Add logging, traces, metrics, and decision records that make the workflow diagnosable.
• Create tests and evaluation cases that measure whether the complete workflow behaves correctly—not merely whether an answer sounds fluent.
• Protect sensitive information and participate in code, design, failure-analysis, and release-readiness reviews.
• Explain your implementation and its trade-offs clearly to engineers, delivery leads, and client stakeholders.

At Level 1, you will not be expected to define the entire client architecture alone. You will be expected to own your assigned module, ask precise questions, surface risks early, and bring it to a production-ready standard with senior review.

What We Are Looking For

Essential Experience

• Approximately 3–6 years of hands-on backend or application engineering experience.
• Evidence that you have built or materially owned production APIs, services, data flows, integrations, or workflow-heavy applications.
• Strong programming ability in JavaScript/TypeScript, Python, Java, C#, Go, or a comparable backend language. Our preference is Node.js/TypeScript, but engineering depth matters more than language loyalty.
• Practical understanding of API design, databases, asynchronous processing, authentication and authorisation, testing, and deployment.
• Ability to reason clearly about state, retries, idempotency, concurrency, permissions, audit trails, and failure recovery.
• Experience debugging real production behaviour rather than only building greenfield demonstrations.
• Clear written and verbal communication, including the ability to explain technical trade-offs without hiding behind framework terminology.

Useful, but Not Mandatory

• Experience integrating LLM or machine-learning capabilities into an application.
• Experience with event-driven systems, state machines, workflow engines, observability, evaluation harnesses, or production incident analysis.
• Experience in healthcare, life sciences, clinical trials, or another regulated or audit-sensitive environment.

Healthcare or clinical-trials experience is preferred, not required at Level 1. We would rather hire a strong production engineer who can learn the domain than a candidate who knows the vocabulary but cannot build dependable systems.

What Does Not Qualify by Itself

Any of the following may be useful experience, but none is sufficient on its own:

• A chatbot or "chat with your documents" application.
• A RAG demonstration built primarily by connecting frameworks.
• A list of AI tools, model names, courses, certificates, or prompt-engineering techniques.
• An application that works in a demo but has no clear handling of permissions, failures, evaluation, or production operations.

How We Will Assess Fit

Our evaluation is designed to identify engineering judgment without asking you to build unpaid project work. We will focus on:

• A structured discussion of a production system you have worked on.
• Questions grounded in your own CV and claimed experience.
• A realistic workflow scenario covering system boundaries, evidence, controls, failure modes, and testing.
• Backend fundamentals and how you would apply them to an AI-enabled workflow.

You are NOT expected to have built a clinical-trials agent before applying.

What Success Looks Like

Within your first six months, you should be able to:

• Implement a bounded agent-workflow module from an agreed design and take responsibility for its quality.
• Integrate models and tools without allowing probabilistic output to bypass deterministic controls.
• Produce traces, tests, and evaluation evidence that support a release decision.
• Diagnose failures across application, tool, data, and model boundaries with progressively less supervision.
• Explain the workflow to a client or domain stakeholder in clear operational language.

Consistent performance at this level creates a path towards Level 2 — Independent Delivery Engineer, where you own a production workflow end to end, including discovery, architecture, integrations, evaluation, failure handling, and release readiness.

Why You Should Join Us

• Build the real systems behind agentic AI. You will work beyond demos and wrappers on the engineering problems that determine whether an AI workflow can operate in production.
• Develop a scarce, durable specialisation. Sentiaflow offers a defined progression across application engineering, platform and evaluation, and applied AI—not a vague instruction to "learn AI."
• Work on consequential operational problems. Clinical-trials workflows demand evidence, traceability, human accountability, and reliable execution.
• Learn through delivery with experienced review. You will own meaningful engineering work while receiving architectural and domain guidance appropriate to your level.
• Influence how the discipline is built. We are at an early stage, so strong engineers will help shape our methods, reusable assets, quality standards, and engineering culture.

We are looking for a talented and driven Data Scientist to join our growing Analytics team in India. In this role, you will work at the intersection of advanced machine learning, scalable MLOps infrastructure, and domain-specific healthcare analytics. You will collaborate closely with cross-functional teams to build, deploy, and maintain production-grade ML models that drive real-world impact in clinical trials and healthcare operations.

KEY RESPONSIBILITIES

End-to-End ML Development

•     Design, build, and optimize predictive models across the full ML lifecycle—from data ingestion to model serving.

•     Conduct rigorous Exploratory Data Analysis (EDA) to surface insights and drive feature engineering decisions.

•     Validate model performance using appropriate statistical techniques and domain knowledge.

MLOps & Production Deployment

•     Deploy, monitor, and maintain production-grade ML models using Databricks MLFlow endpoints and Unity Catalog.

•     Implement CI/CD pipelines for model versioning, experiment tracking, and automated retraining.

•     Ensure model reliability, observability, and performance in live production environments.

Language Models & LLM Applications

•     Apply transformer-based models (BERT, ClinicalBERT, Trial2Vec) for NLP tasks including classification, NER, and information extraction.

•     Build and maintain vector similarity search pipelines for semantic retrieval and recommendation use cases.

•     Fine-tune pre-trained models for domain-specific applications in clinical and healthcare contexts.

•     Support exploratory work around LLM integration and prompt engineering for internal tooling.

Domain-Driven Analytics

•     Apply advanced analytics within complex healthcare and clinical trial datasets—including patient records, trial protocols, and adverse event data.

•     Translate ambiguous business problems into structured analytical frameworks with measurable outcomes.

•     Partner with domain experts, product managers, and engineering teams to deliver data-driven solutions.

REQUIRED QUALIFICATIONS

Education

•     Bachelor’s or Master’s degree in Computer Science, Statistics, Mathematics, Bioinformatics, or a closely related field.

Experience

•     2–4 years of hands-on experience in a data science or machine learning role.

•     Demonstrable experience deploying ML models in production environments (not just prototyping).

Technical Skills

•     Strong proficiency in Python (pandas, NumPy, scikit-learn, PyTorch / TensorFlow).

•     Experience with Databricks, MLFlow (experiment tracking, model registry, endpoints), and Unity Catalog.

•     Hands-on experience with BERT-family models and Hugging Face Transformers library.

•     Familiarity with vector databases (e.g., FAISS, Pinecone, Weaviate) and embedding-based retrieval.

•     Solid understanding of SQL and working with large structured/unstructured datasets.

•     Exposure to cloud platforms (AWS / GCP / Azure) and distributed computing frameworks (Spark).

GOOD TO HAVE

•     Prior experience with clinical trial data standards (CDISC, CDASH, SDTM) or healthcare ontologies (SNOMED, ICD-10).

•     Familiarity with Trial2Vec or similar trial-to-vector embedding approaches.

•     Experience with LLM fine-tuning, RAG pipelines, or prompt engineering in a production setting.

•     Knowledge of regulatory and compliance considerations in healthcare AI (e.g., FDA guidelines, HIPAA).

•     Contributions to open-source ML projects or published research.

THIS ROLE IS NOT FOR YOU IF…

•     You have strong SQL/BI skills but limited hands-on ML modelling experience — or you’ve built models only in notebooks without ever deploying them to production.

•     Your LLM exposure is limited to API calls and prompt engineering — with no experience fine-tuning models, working with embeddings, or building vector search pipelines.