Selected engineering work

Agentic AI + retrieval

AutoCare Agent

A six-node LangGraph diagnostic workflow connecting hybrid retrieval, four tools and grounded decision support.

PythonLangGraphFastAPIOllamaQwen2.5 3BOpenSearchSentence-TransformersPostgreSQLSQLAlchemyAlembicRedisCeleryStreamlitJWTDocker ComposepytestGitHub Actions
Choose your depth
Architecture, choices & results

Combine service-manual retrieval, service history and diagnostic tools for grounded vehicle-diagnostic decision support.

Follow the flow from input to evaluated output. Select a stage to inspect its role.

Validation — Validate the request.

FastAPIPostgreSQLOpenSearchRedisCeleryOllamaStreamlit
Engineering decision

Why hybrid retrieval?

Problem
Ground vehicle-diagnostic responses using unstructured service manuals and structured service history.
Approach
Combine BM25 lexical search with dense semantic retrieval.
Trade-off / scope
Separate lexical-only and dense-only baseline scores are not supplied.
direct evidenceRetrieval evaluation

Hybrid BM25 + all-MiniLM-L6-v2 dense retrieval evaluated across 100 diagnostic queries.

Documented result90% Recall@5100-query retrieval evaluation (local research; completed): Recall@5 for hybrid retrieval. 100 diagnostic queries; 20 service manuals; approximately 4,500 document chunks. Recall@5: 90%. Hybrid retrieval over 100 diagnostic queries; profile-supported, evaluation artifacts pending. Indexed document chunks: ~4,500. Approximate corpus size: 20 service manuals. Service manuals: 20. Indexed automotive service manuals. Decision-support POC, not a substitute for a certified technician. Recall describes the 100-query retrieval evaluation, not field diagnostic accuracy. Separate lexical/dense scores and scenario pass rates are not supplied.
supporting evidenceSystem architecture

LangGraph connects request validation, planning, retrieval, tools, safety checks and response generation.

Documented resultSix-node agentic workflowDiagnostic scenario validation (unspecified; completed): Scenario testing; detailed outcomes not supplied. 40 vehicle-diagnostic scenarios; 1,000 PostgreSQL service-history records. Diagnostic scenarios tested: 40. Count of tested scenarios, not a scenario success rate. Service-history records: 1,000. PostgreSQL service-history dataset size; not number of successful diagnoses. Decision-support POC, not a substitute for a certified technician. Recall describes the 100-query retrieval evaluation, not field diagnostic accuracy. Separate lexical/dense scores and scenario pass rates are not supplied.
supporting evidenceEngineering decision

Combine BM25 lexical search with dense semantic retrieval.

Documented resultWhy hybrid retrieval?100-query retrieval evaluation (local research; completed): Recall@5 for hybrid retrieval. 100 diagnostic queries; 20 service manuals; approximately 4,500 document chunks. Recall@5: 90%. Hybrid retrieval over 100 diagnostic queries; profile-supported, evaluation artifacts pending. Decision-support POC, not a substitute for a certified technician. Recall describes the 100-query retrieval evaluation, not field diagnostic accuracy. Separate lexical/dense scores and scenario pass rates are not supplied.

100-query retrieval evaluation: completed. 100 diagnostic queries; 20 service manuals; approximately 4,500 document chunks. Diagnostic scenario validation: completed. 40 vehicle-diagnostic scenarios; 1,000 PostgreSQL service-history records.

A documented safety and scope boundary.

Dedicated safety-validation stage; diagnostic decision support does not replace certified technicians. Decision-support POC, not a substitute for a certified technician. Recall describes the 100-query retrieval evaluation, not field diagnostic accuracy. Separate lexical/dense scores and scenario pass rates are not supplied.

Documented services

FastAPI · PostgreSQL · OpenSearch · Redis · Celery · Ollama · Streamlit

Context limited to AutoCare

Challenge this project.

Ask about the decisions, evaluation or boundaries behind the work.

Deterministic answers from documented portfolio evidence · project-scoped

A claim is only the beginning

Where did I use it?
What did I build?
How did I measure it?

Follow the projects, decisions and evaluation behind each answer. Every important claim has somewhere to go.

Direct · measured implementation Supporting · documented use Related · adjacent experience
An inspectable claimPG / 001
CAPABILITYRetrieval-augmented generation
IMPLEMENTATION

BM25+semantic retrieval

Trace the evaluation behind the number