Immigration services · built 0 to 1
EB1A Experts platform
AI platform that drafts, verifies and scores EB1A petitions, built from scratch and run in production on AWS.
1,000+ clients · 70% less manual work
Multi-agentRAGMCPFastAPIAWS EKSKongGrafanaSpaCyBERT
Overview
I led the build from the first commit to production in about a year. The platform generates profile content, verifies it, and benchmarks each applicant against historical successful cases. It then gives criterion-by-criterion approval odds and recommendations.
Architecture
FastAPI microservices on Kubernetes (EKS) behind Kong for auth and rate limiting, with the Grafana stack for metrics, traces and logs.
- Kong API Gateway: auth, rate limiting, routing
- Generation service: multiple LLM providers
- Verification service: lexical rules (NLTK) and NER (SpaCy, BERT)
- Recommendation engine: multi-agent RAG with web search and MCP tools
- Observability: Grafana, Prometheus, Jaeger, Loki
- Security: Trivy image scans, SonarQube
Client → Kong → EKS services → LLMs / NLP models → verification → scored recommendations
What I did
- Owned the backend architecture, CRUD APIs and AI services
- Built the multi-agent recommendation engine and per-criterion scoring
- Built content verification with NER and custom rules
- Set up CI/CD with GitHub Actions, Kong, EKS and the observability stack
- Interviewed and mentored engineers joining the team
References
- Kong Gateway
API gateway for auth and rate limiting.
- Grafana stack
Metrics, traces (Jaeger) and logs (Loki).