FPT Software AI Center
Senior AI Engineer
Sep 2024 - Present
Provided technical leadership for an 8-9-person team across system design, experimentation and code reviews; clarified stakeholder requirements and coordinated frontend/backend delivery.
Enterprise contract intelligence & agent architecture
- Designed the AI architecture for an enterprise contract-analysis platform serving approximately 1,000 users, using LangGraph to coordinate multi-step document review and tool-based analysis.
- Made retrieved answers traceable to source documents through citation-backed retrieval; added response caching to reuse LLM outputs and evaluated lower-cost models against task requirements.
- Preserved contract metadata and master-data history across revisions through snapshot versioning, supported by PostgreSQL/pgvector and OpenSearch data models.
Enterprise integration & feature access
- Integrated the contract platform with an internal AI service, addressing cross-service authentication and secure connectivity requirements within enterprise infrastructure.
- Designed Salesforce Canvas and SSO integration to embed the contract platform into existing business workflows, defining application embedding and API contracts across systems.
- Enabled business-unit-specific feature access through a Business Line permission matrix, enforcing the same feature restrictions across backend APIs and the application UI.
Knowledge graphs & software lifecycle intelligence
- Built traceable links between requirements, documentation, source code and defects in a knowledge graph, giving QA and engineering a connected view of information spread across Rally, wikis and code repositories.
- Led a feasibility study for test generation grounded in requirement dependencies; modeled epic, feature and user-story relationships to supply context for scenarios spanning multiple requirements.
Document intelligence & applied research
- Designed ingestion to preserve document layout and section hierarchy for contextual retrieval; evaluated vision-language classification and document parsers to inform the pipeline design.
