Lead Principal Agentic RAG Engineer – Python, Delhi | BestKaam
AI Startup
On-siteOn-siteFull-time$220k – $350k / year
About this role
Lead Principal Agentic RAG Engineer – Python, Delhi | BestKaam
Lead Principal Agentic RAG Engineer – Python & AI Platforms
Actively Reviewing
Bridge AI
Delhi Full-Time 4–8 yrs exp Posted 1 month ago · Apply by Aug 4, 2026
Job Description
About the Role
We are hiring a senior, hands-on Agentic RAG Engineer to design, build, and operate the Retrieval-Augmented Generation platforms that power our autonomous AI agents.
This is a lead-by-example role:
- You design the architecture
- You write the Python
- You ship to production
- You mentor engineers by building real systems
You will lead the technical direction of RAG and agent memory systems, while remaining deeply involved in implementation, observability, and operational readiness.
GCP is our primary platform, but all designs should be multi-cloud capable.
Key Responsibilities
RAG & Backend Engineering (Python-First)
- Design and build production-grade RAG pipelines
- Implement:
- Retrieval strategies
- Vector database integrations
- Agent memory and state management
- Prompt orchestration and chaining
- Build scalable Python services using FastAPI / Django / similar
- Integrate LLM APIs (OpenAI, Claude, Gemini) and open-source models (Llama, Mistral)
- Implement model/version rollout, rollback, and simulation testing
Agentic Systems & Workflow Design
- Build and operate multi-step agent workflows
- Enable:
- Tool calling
- Human-in-the-loop interventions
- Safe agent execution patterns
- Define patterns for:
- Prompt versioning
- Context management
- Token and cost control
- Collaborate closely with AgentOps to ensure production-safe execution
Full-Stack & Observability
- Design and contribute to internal UIs for:
- Agent execution monitoring
- Decision and reasoning audits
- Prompt testing and visualization
- Implement structured logging and telemetry for:
- Retrieval quality
- Agent decisions
- Token usage and latency
- Work with Prometheus, Grafana, OpenTelemetry, or ELK-based stacks