Multi-step LLM agentic workflows for candidate matching with tool-calling, structured outputs, and ranked recommendations.
LangGraphLangChainFastAPIChromaDBGroqMCPReact
Problem
Candidate matching needed multi-step reasoning — resume retrieval, tool use, and ranked recommendations — not a single prompt.
Approach / architecture
Built LangGraph/LangChain agentic workflows with tool-calling agents, structured outputs, resume retrieval, and MCP tooling, exposed through a React dashboard.
TOPOLOGY: ASYNCHRONOUS AGENTIC EVALUATION PIPELINELANGGRAPH • VECTOR STORE • EVALUATOR
[Candidate Input]Schema Verified
↓
[Embedding Node]Dense Vector
↓
[Vector Index]Top-k Retrieve
↓
[Agentic Evaluator]Deterministic Graph
↓
[Scored Matrix]Ranked output
STAGE 01
[Candidate Input]
Resume / Spec JSON
Schema Verified
▶flow
STAGE 02
[Embedding Node]
text-embedding pipeline
Dense Vector
▶flow
STAGE 03
[Vector Index]
HNSW nearest filter
Top-k Retrieve
▶flow
STAGE 04 :: LLM AGENT
[Agentic Evaluator]
LangGraph re-ranking chain
Deterministic Graph
▶flow
STAGE 05
[Scored Matrix]
JSON confidence matrix
Ranked output
Multi-step agentic workflow with retrieval, evaluation, and ranked structured output
Outcome
End-to-end agentic matching pipeline with ranked recommendations and dashboard visibility for operators.