Custom Software Engineering Lead

Project Role : Custom Software Engineering Lead

Project Role Description : Own the technical direction and architecture of custom software solutions, leading teams through design and delivery. Set development standards and ensure code quality, scalability, and performance aligned to business objectives.

Must have skills : AI Agents & Workflow Integration

Good to have skills : AI & Data Solution Architecture

Minimum 7.5 Year(s) Of Experience Is Required

Educational Qualification : 15 years full time education

About The Role

Platforms value is in its agents. Eight specialist services — five diagnostic and three planning — each call a combination of graph traversal, vector search, MCP tool invocation, and Bedrock reasoning to answer questions no human could answer quickly from raw a alone.

This engineer builds Core, all eight agent services, the MCP Gateway, and the retrieval pipeline. The role that turns the knowledge graph into actionable intelligence. A second engineer joins at Week 9 when Phase 1 agent work opens in parallel with infrastructure.

Responsibilities

Build Core: the 13-node LangGraph StateGraph, hybrid BM25+vector retrieval pipeline, context packer with budget and provenance tracking, and Bedrock model router

Implement all 8 agent services as FastAPI microservices: APM/Change Log, Historic RCA, Cloud Health, Integration Impacts, Log Analyzer, Change Impact, Designer, and Story Planner

Build the Connector SDK base contract and migrate all 9 connectors to it (GitHub, Confluence, ServiceNow, JIRA, CMDB, SonarQube, Dynatrace, REMC, RepoMapping)

Build and own the MCP Gateway: per-agent RBAC, request routing, rate limiting, token budget enforcement, and Postgres audit log of every tool call

Implement HITL approval gates in both MI and CR orchestrators using LangGraph wire blast-radius threshold config and Slack approval shortcut

Complete the MCP server set: Splunk MCP, Elasticsearch MCP, Slack MCP, and GoAlert write client

Build and maintain the RAGAS evaluation harness as a continuous job instrument confidence-provenance vectors on every agent output

Implement idempotent OpenCypher MERGE patterns, provenance flagging, and the freshness ledger

Wire entity resolution output into the master index so all agent queries resolve to canonical IDs

Contribute to the Phase 3 learning loop: agent-inferred edge proposal queue, human valiion UI, and graph write-back

Required Skills

LangGraph and LangChain: multi-node stateful StateGraph design, conditional routing, HITL gate patterns, retry and fallback logic

Python: async FastAPI service design, Pydantic models, production-grade error handling and structured logging

LLM APIs: Amazon Bedrock (Claude Haiku, Sonnet, Titan v2), prompt engineering, token budget management, streaming responses

Graph abases: OpenCypher queries against Neptune, idempotent MERGE patterns, traversal for blast-radius analysis

Vector search: OpenSearch k-NN, Titan v2 embeddings, hybrid BM25+dense retrieval, relevance evaluation

MCP (Model Context Protocol): server and client implementation, stdio and HTTP transport, tool schema design

RAG evaluation: RAGAS metrics, confidence scoring, provenance tracking, retrieval quality measurement

API integration: REST, OAuth 2.0, webhook receivers, rate limit handling across enterprise systems

Testing: unit and integration test patterns for LLM-dependent code deterministic test harness design

Observability: OTel SDK instrumentation with trace-ID propagation across agent hops