Affan KhamseAI Engineer

New York, USA

I own the credentialing system end to end. Clinicians must pass license, sanction and health checks before they start work. My agent runs those checks, approves what it can prove, and sends the rest to a credentialer. It processes 1,000+ documents a week across 70+ onboardings.

• Primary source verificationVerifies licenses, sanctions and registries against 50+ official sources in parallel. Each check went from about 5 minutes of manual work to under 30 seconds.
• The agent proposes, the code decidesThe agent interprets messy checklist lines. Code runs only registered checks. A credentialer approves a new interpretation once, and it becomes reusable policy.
• Built where users already workCredentialers review results in a Chrome extension inside JobDiva. 12 pilot users run it today, and their feedback drives the fixes.
• Layla, the onboarders' assistantAn AI assistant onboarders use in production. She drafts reminders, candidates and placements, and remembers how each person likes to work. The user confirms every action.

• Python
• TypeScript
• Next.js
• Vercel AI SDK
• Gemini
• Google Document AI
• GCP Cloud Run
• Cloud SQL
• Agentic Systems

Do not rip out the system a company has years of data in. Put an agent on top of it.

New York, USA

Owned the retrieval and generation path for an AI call assistant surfacing live suggestions mid-conversation. Every millisecond of lag was a rep waiting on the model.

• Parallelized retrievalConcurrent multi-source lookups instead of sequential fetches
• Hierarchical chunkingLayout-aware segmentation that preserved document structure
• Hybrid searchDense embeddings combined with BM25 keyword matching
• Prompt compressionContext trimmed from 4,900 to 1,950 tokens per call

• AWS Lambda
• DynamoDB
• AWS SAM
• AssemblyAI
• AWS Bedrock
• WebSockets
• FastAPI
• JWT
• RBAC
• Redis
• CI/CD
• Serverless Architecture

New York, USA

Built an AI assistant platform end-to-end as one of three engineers, from semantic retrieval to the async backend serving it.

• Developed an AI assistant platform using LangChain, OpenAI Embeddings, and Chroma vector store, achieving accurate document retrieval on 500+ internal test queries with semantic search
• Redesigned backend by migrating Flask to FastAPI with asynchronous processing and optimized database queries, reducing API latency by 30% through performance testing
• Led Agile development in a 3-person team, using Azure DevOps to manage sprints and implementing CI/CD pipelines for automated testing and deployment to AWS

• FastAPI
• LangChain
• OpenAI
• Chroma
• Vector Database
• Semantic Search
• Azure DevOps
• Flask
• PostgreSQL
• Docker
• RESTful APIs
• Async Processing
• CI/CD