Agentic systems
Multi-agent orchestration with LangGraph: parallel and conditional flows, MCP tool layers, checkpointing, human-in-the-loop and voice interfaces.
AI Engineer · Noida, India
I build AI agents and analytics that hold up in production.
I'm Pratham Rana, an AI Engineer at Polestar Analytics. For 2.5 years I've designed multi-agent platforms with LangGraph and MCP, RAG pipelines and NL-to-SQL analytics on Azure and AWS, used by 500+ business users across healthcare and retail.
2.5 yrsbuilding enterprise GenAI
500+business users on platforms I built
<2sP95 latency on real-time analytics
40%smaller analytics backlog for a biopharma client
01 / About
Where language models meet real business data
At Polestar Analytics I turn business requirements into LLM-powered platforms for healthcare and retail clients: agents that plan and call tools, retrieval systems that ground answers in company knowledge, and conversational analytics that let people ask their data questions in plain English.
Most of my effort goes into the parts that make AI dependable in production: guardrails and PII redaction, tracing and evaluation, latency budgets, and knowing exactly what each agent costs to run. I ship on FastAPI, Docker and Azure Kubernetes Service.
- Role
- AI Engineer, Polestar Analytics
- Based in
- Noida, India (IST)
- Focus
- Agentic AI · RAG · LLMOps
- Core stack
- Python · FastAPI · LangGraph
- Cloud
- Azure · AWS
- Certified
- Databricks GenAI Engineer
RAG & conversational analytics
Hybrid search with reranking, and NL-to-SQL over Snowflake and Databricks so business users can ask follow-up questions of their own data.
LLMOps & governance
Evaluation with RAGAS and LLM-as-a-judge, tracing in LangSmith and Langfuse, guardrails, PII redaction and per-agent cost tracking.
02 / Experience
Experience & education
AI Engineer · Polestar Analytics
- Designed and delivered enterprise AI platforms for healthcare and retail clients, turning business requirements into scalable LLM-powered applications and analytics workflows used by 500+ business users.
- Engineered agentic systems with LangGraph, LangChain, FastAPI and enterprise LLM providers: MCP-based tool execution, sandboxed tools, guardrails, PII detection, agent FinOps, voice interaction, conditional routing, state management and observability, deployed on Docker and AKS.
- Built conversational analytics and NL-to-SQL capabilities over enterprise data platforms, so business users can explore data, ask contextual follow-up questions and act on the insights.
Certification
Databricks Certified Generative AI Engineer Associate
Education
B.Tech, Information Technology
03 / Projects
Selected work
Production platforms I've designed and built for enterprise teams.
Polestar Analytics
Agenthood
Multi-agent workflow orchestration platform
A platform for building, running and governing multi-agent workflows across the enterprise. I architected its backend orchestration layer.
- Graph-based execution with FastAPI and LangGraph: sequential and parallel agent tasks, conditional branching and state transitions.
- MCP servers as a standard tool layer, so agents discover and invoke enterprise tools, data sources and APIs through function calling.
- Guardrails and PII detection and redaction on inputs, tool calls and outputs, enforcing content policy and data privacy.
- Agent FinOps: per-agent token metering, cost attribution by model and provider, budgets with alerts, and actual-vs-forecast spend dashboards.
- Real-time voice chat with agents through speech-to-text and text-to-speech pipelines, alongside text channels.
Retail client
AI Insight Portal
Decision intelligence platform
Monitors business KPIs across enterprise retail datasets and surfaces what changed, how much it matters and why.
- Configurable signal framework for KPI monitoring, threshold evaluation, severity scoring and signal execution.
- Automated anomaly and variance analysis that detects unusual KPI movements and triggers downstream insight generation.
Biopharmaceutical client
Healthcare Conversational Intelligence
Ask-your-data analytics over Snowflake
LLM-driven conversational workflows, built with LangChain and FastAPI, that turn business questions into contextual analytical requests over enterprise healthcare data.
<2sP95 latency under high concurrency
−40%analytics backlog
More case studies on the way
Want the deeper story behind any of these? Ask me on a call.
04 / Skills
Toolkit
Agentic AI
- AI Agents
- Multi-Agent Systems
- Agent Orchestration
- Model Context Protocol (MCP)
- MCP Servers & Clients
- Agent-to-Agent (A2A)
- OpenAI Agents SDK
- CrewAI
- AutoGen
- Agent Memory
- Planning & Reasoning
- Tool Binding
- Sandboxed Tool Execution
- Voice Agents (STT/TTS)
- Agent FinOps
- Conditional Routing
- Checkpointing
- Human-in-the-Loop
- Guardrails
- SSE
- Socket.IO
Generative AI & LLMs
- Azure OpenAI
- OpenAI API
- Anthropic Claude API
- Google Gemini
- Hugging Face Transformers
- PyTorch
- LangChain
- LangGraph
- LlamaIndex
- Prompt Engineering
- Context Engineering
- Structured Outputs
- Function / Tool Calling
- Fine-Tuning (LoRA, PEFT)
- Multimodal LLMs
RAG & Retrieval
- RAG
- Agentic RAG
- Embeddings
- SentenceTransformers
- Vector Search
- Hybrid Search
- Semantic Search
- Reranking
- Chunking Strategies
- Knowledge Graphs
- Semantic Caching
LLMOps & Evaluation
- LLM Evaluation
- LLM-as-a-Judge
- RAGAS
- LangSmith
- Langfuse
- Observability & Tracing
- Prompt Versioning
- A/B Testing
- Token & Cost Optimization
- Latency Optimization
- Responsible AI
- PII Detection & Redaction
- Hallucination Mitigation
Cloud & MLOps
- Azure AI Foundry
- Azure AI Search
- Azure Key Vault
- Azure Blob Storage
- AWS Bedrock
- AWS S3
- IAM
- KMS
- Secrets Manager
- Docker
- Kubernetes
- AKS
- CI/CD
- GitHub Actions
- Linux
Programming & Backend
- Python
- SQL
- FastAPI
- SQLAlchemy
- asyncio
- Pydantic
- Uvicorn
- REST APIs
- Streaming APIs
- WebSockets
- Git
- pytest
Databases & Vector Stores
- PostgreSQL
- Snowflake
- Databricks SQL
- Azure SQL
- PgVector
- Qdrant
- Pinecone
- FAISS
- Chroma
- Redis
Enterprise Integrations
- Microsoft Teams Bot Framework
- Adaptive Cards
- Microsoft Graph API
- OAuth2
- MSAL
- JWT
- Role-Based Access Control
- Microsoft Copilot Studio
05 / Contact
Let's talk
Hiring for an AI engineering role, scoping an agentic or RAG build, or want to compare notes on running LLMs in production? Pick a 30-minute slot and it goes straight onto my calendar.
- 30 minutes
- Video call
- Times shown in your time zone
Book a 30-minute call
Send me a short note with what you'd like to discuss and a few times that suit you. I'll reply with a calendar invite.
Request a meeting