Selected work
My projects
A quick overview of the business problems, solution approaches and technologies.
Project 01 • AI / RAG
AI-Assisted Visa Application Screening
Client described: Federal Government of Canada | Role: GenAI / Backend Engineer
Problem: Large volumes of study visa, work visa, business visa, permanent residency and citizenship applications made manual document review time-consuming.
Solution: Our team built an AI-assisted screening workflow that processes application documents, retrieves relevant eligibility guidelines using RAG, identifies missing information or inconsistencies, and generates structured summaries for officers. Authorized officers remain responsible for final decisions.
How the workflow works
01
SubmitApplicant uploads forms and supporting documents.
02
ExtractDocument processing extracts and organizes relevant information.
03
Retrieve & analyzeRAG finds relevant policies; an LLM helps flag gaps and inconsistencies.
04
ReviewOfficer receives an evidence-based summary for human assessment.
My work: Python backend development, API integration, document processing, RAG pipeline integration, prompt design, error handling and testing.
Relevant technologies:
PythonFastAPIRAGLLMsEmbeddingsVector SearchDocument Processing
The exact LLM, vector database, OCR service and cloud services used in this project have not yet been confirmed. Add only the technologies you personally used before presenting this as a technical case study.
Project 02 • Enterprise AI
RAG-Based Intelligent Commerce Assistant — PepsiCo
Developed a natural-language interface for enterprise Inventory and Location data. Instead of manually constructing GraphQL queries, business users could ask questions in plain English. The solution used Claude LLM to interpret requests, generated GraphQL queries, called the relevant data APIs and returned concise responses.
My work: Claude integration, MCP server with dynamic tool registration, FastAPI endpoints, session context, multi-domain orchestration, Okta JWT authentication, logging and Docker packaging.
PythonFastAPIClaudeAzure DatabricksGraphQLMCPOkta JWTDocker
Project 03 • Recommendations
Conversational AI Recommendation System
Built a personalized product discovery experience combining behavioral analytics, recommendation models, vector similarity search and LLM-driven conversations. Users could describe their preferences naturally and receive relevant product suggestions.
My work: Recommendation APIs, embedding pipelines, semantic search, RAG integration, feature processing, ranking optimization, A/B testing and deployment support.
PythonFastAPIMySQLFAISS / PineconeRAGAzureDocker