Python - Agentic AI Developer

Python Developer — Agentic AI Platform
Healthcare RCM Automation | Calpion Inc.
Location
Experience
Positions
Bangalore, India
Reports to
4–8 years
13
About the Role
PM & Technical Lead
We are building an enterprise agentic AI platform that serves as the automation backbone
across three healthcare RCM product lines — Artemis (ABA therapy management), Olympus.
This platform replaces traditional RPA with intelligent agents that reason, adapt, and self -heal —
combining deterministic automation with LLM -powered decision -making for healthcare
workflows.
You will be a core member of the platform team, responsible for building reusable agent
workflows, MCP tool integrations, and the LLM reasoning layer that powers denial management,
payer intelligence, document processing, and browser automation across all products.
This is not a chatbot or copilot role. You will be building production agents that process
healthcare claims, interact with payer portals, generate appeal letters, parse EDI transactions,
and make decisions that directly impact revenue recovery. The work requires equal comfort with
infrastructure -level Python, LLM prompt engineering, and healthcare domain understanding.
What You Will Work On
Agent Development & Orchestration
• Design and build agentic workflows using LangGraph for complex, multi -step healthcare
processes — denial management pipelines, eligibility verification chains, prior
authorization workflows
• Implement the 4 -tier cost architecture: deterministic scripts (Tier 1), LLM fallback for self
healing (Tier 2), cached AI navigation (Tier 3), and full reasoning agents (Tier 4)
• Build multi -agent systems where specialized agents (claim analyzer, appeal writer, payer
policy checker) collaborate through structured handoffs
• Integrate agents with AWS Step Functions for durable, stateful execution of long -running
workflows with human -in -the -loop approval gates
MCP Tool Development
Page 1
Calpion Inc. | Confidential
• Build reusable MCP (Model Context Protocol) server tools wrapping existing APIs —
AdvancedMD, CollaborateMD, clearinghouse APIs (Waystar, Change Healthcare,
Availity), payer portal integrations
• Design the MCP tool library architecture so that every connector built for one product.
• Implement custom MCP tools for healthcare -specific operations: EDI 835/837 parsing,
CARC/RARC code lookup, CPT/ICD -10 reference queries, payer policy retrieval
Browser Automation
• Build and maintain Playwright -based browser automation for healthcare portals
(AdvancedMD, payer portals, clearinghouse portals)
• Implement the LLM fallback self -healing pattern — deterministic Playwright as primary
path, Claude Vision via Bedrock for automatic recovery when portal UI selectors break
• Develop cached navigation plans for payer portals — LLM discovers navigation once,
caches the plan, replays deterministically on subsequent runs
• Integrate with Amazon Bedrock AgentCore Browser Tool for managed, isolated browser
sessions
Must -Have Skills
Python (Core)
• 4+ years production Python — deployed services handling real workloads, not scripts or
notebooks
• Strong async Python (asyncio, aiohttp) — browser automation and agent loops are
inherently async
• Pydantic for data validation and structured outputs
• Experience with at least one web automation framework: Playwright (strongly preferred),
Selenium, or Puppeteer via Pyppeteer
• Comfortable with Python packaging, dependency management, and writing maintainable
code
AWS Services
• Hands -on experience with at least 4 of: ECS Fargate, Lambda, Step Functions, SQS,
SNS, EventBridge, S3, RDS PostgreSQL, Secrets Manager, Bedrock
• Understanding of IAM policies, VPC networking, and security groups
• Experience with CDK or CloudFormation for infrastructure -as -code (CDK TypeScript
preferred given existing codebase)
Page 2
LLM / AI Engineering
Calpion Inc. | Confidential
• Practical experience building applications with LLM APIs — Claude (Anthropic), GPT
(OpenAI), or equivalent
• Understanding of prompt engineering patterns: ReAct (Reason + Act), chain -of -thought,
few -shot examples, structured output extraction
• Experience with at least one agent framework: LangGraph, LangChain, CrewAI, or
AutoGen
• Understanding of RAG — vector embeddings, similarity search, knowledge base
construction
• Awareness of token economics — knowing when to use an LLM and when a regex or
SQL query is the right answer
Software Engineering Fundamentals
• Git workflows, code review practices, CI/CD pipelines
• REST API design and consumption
• PostgreSQL — complex queries, indexes, JSONB columns
• Error handling for unreliable external systems (retry logic, circuit breakers, dead letter
queues)
• Structured logging, CloudWatch integration, distributed tracing concepts
Good -to -Have Skills
• Healthcare RCM domain knowledge — EDI 837/835 formats, CARC/RARC denial
codes, CPT/HCPCS coding, clearinghouse workflows, payer portal navigation
• MCP (Model Context Protocol) — experience building MCP servers/tools or familiarity
with the specification
• Amazon Bedrock — AgentCore, Knowledge Bases, or model invocation via Bedrock
Runtime API
• LangGraph specifically — graph -based state machine design for multi -step agent
workflows
• Document AI — PDF extraction, OCR, table extraction from scanned documents
• HIPAA compliance awareness — PHI handling, encryption requirements, audit trail
needs
• Playwright MCP server for browser automation via AI agents
• Voice/telephony integration — Twilio, Deepgram, ElevenLabs (relevant for Olympus
voice bot)
• Prior experience replacing or working alongside RPA tools (UiPath, Automation
Anywhere, Blue Prism)
Page 3
What We Are NOT Looking For
Calpion Inc. | Confidential
• Pure data scientists or ML researchers — this is an engineering role, not a modeling
role. We use frontier LLMs via API, we don’t train models.
• Chatbot/copilot builders — our agents take real actions on production systems (submit
claims, navigate portals, post payments). This requires a production engineering
mindset, not a demo mindset.
• Framework tourists — we need people who can evaluate when to use LangGraph vs a
simple Python function, when to use an LLM vs a regex, when to use an agent vs a cron
job. Tool selection judgment matters more than tool familiarity.
Page 4