JOB DETAILS:

Role: Senior AI Engineer

Location: India (Remote)

Type of Hiring: Full Time

:

Senior AI Engineer— Agentic Systems

Agentic Architecture · MCP · Langchain · RAG · Guardrails · Human-in-the-Loop · Python · FastAPI · AWS Agent Core

Experience 6 – 9 years hands-on

Location Remote

Role Individual contributor — you code daily. No people management.

Core stack Agentic Architecture, MCP, Langchain, RAG, Guardrails, Human-in-the-Loop, Python, FastAPI, AWS, AWS Agent Core THE ROLE

Design and run agentic AI systems that are safe to put in front of real users — agents that plan, call tools, know when to stop, and know when to ask a human. This is the deep-AI seat on the team: architecture, guardrails, MCP connectivity and human oversight are the job, not a side project.

WHAT YOU WILL DO

▪ Design agentic architectures: planning and reasoning loops, task decomposition, multi-agent orchestration, shared state and memory, deterministic control flow with Langchain.

▪ Build and operate MCP (Model Context Protocol) servers, clients and connectors — expose internal systems as tools and resources with clean schemas, auth, versioning and lifecycle management.

▪ Design tool interfaces agents can use reliably: typed schemas, timeouts, retries, partial-failure handling, cost and action ceilings.

▪ Implement guardrails across the lifecycle — input validation, prompt-injection and jailbreak defence, PII redaction, toxicity and topical constraints, grounding and hallucination checks, output schema enforcement, tool-call authorisation.

▪ Design human-in-the-loop: approval gates before consequential actions, confidence thresholds and escalation, review queues, interrupt and resume of long-running runs, and feedback that flows back into evaluation.

▪ Build RAG that agents can rely on — ingestion, chunking,



hybrid retrieval, re-ranking, grounding and citations.

▪ Build agent evaluation and observability: trajectory and tool-call accuracy, task success rate, token and cost telemetry, tracing (LangSmith, Langfuse, OpenTelemetry), red-team and regression suites.

▪ Ship and run it: production Python/FastAPI services on AWS (Lambda, ECS/EKS, ECR, S3, Redis) with IaC, CI/CD and autoscaling.

MUST HAVE

Agentic AI & MCP — primary

▪ Shipped at least one agentic system beyond a prototype — tool use, multi-step planning, real failure handling, real users.

▪ LangChain or equivalent orchestration (LangGraph, CrewAI, AutoGen, Strands, Bedrock Agents): state machines, checkpointing, interrupts, resumption.

▪ Hands-on MCP: building or integrating MCP servers, clients and connectors to internal systems and third-party tools.

▪ Multi-agent patterns —and the judgement to know when one well-scoped agent is the better answer.

▪ Prompt engineering, structured output, function/tool calling, context and memory management.

Guardrails & human-in-the-loop — primary

▪ Guardrails implemented in production: injection defence, PII redaction, content filtering, grounding checks, output schema enforcement, tool authorisation.

▪ Hands-on with at least two of: AWS Bedrock Guardrails— or an equivalent in-house control layer.

▪ Human-in-the-loop design: approval gates, escalation thresholds, review queues,



reviewer UX and feedback loops — and explicit thinking on where automation should stop.

▪ Evaluation discipline: golden datasets, trajectory and tool-call evaluation, LLM-as-judge, regression suites, human review.

RAG & LLMs

▪ Production RAG: embeddings, vector stores (pgvector, OpenSearch, Pinecone, FAISS, Chroma)

▪ Current model landscape — Claude, OpenAI, AWS Bedrock

Engineering & cloud

▪ 6–9 yrs hands-on engineering; 4+ yrs production Python (3.10+); solid FastAPI (Pydantic v2, async, auth, OpenAPI).

▪ AWS: Lambda, S3, ECR, ECS, ElastiCache for Redis, Kubernetes (EKS); Docker; CI/CD; observability.

Production ownership

▪ CI/CD, infrastructure-as-code, containerised builds, blue/green and canary releases, tested rollbacks.

▪ Zero-downtime releases: backward-compatible contracts, safe schema migrations, graceful degradation.

▪ Logging, metrics, tracing, dashboards and alerts against meaningful SLOs.

▪ Guardrail config, policy and evaluation suites versioned and shipped through the same pipeline as the code.

Communication & application design — non-negotiable

▪ Take an ambiguous problem to a documented design — data flow, API contracts, failure modes, scale and cost trade-offs — and defend it in review.

▪ Write clearly: design docs, PR descriptions, incident write-ups.

▪ Explain trade-offs to product, design and business stakeholders; disagree constructively.

GOOD TO HAVE

▪ Published or maintained an MCP server; contributions to agent frameworks.

▪ Graph RAG and knowledge graphs; computer-use or browser agents.

Hands-on means hands-on

You will design, write, review and ship production code every week, and own how it behaves in production. Coordination-only, delivery-management or architecture-only profiles are not a fit.

📌 Senior Artificial Intelligence Engineer (India)
🏢 PrithviTech Global|India| US| UAE| Netherlands
📍 India