Fabio Pecora

About

Early-career full-stack builder who created an AI interview-prep platform used by 200+ students, tested it with a 25-student study group, and earned recognition through the CUNY Startup accelerator, including prize money and local college adoption. Has also shipped compliance-sensitive AI products in healthcare marketing and operational tools like invoice approval systems, showing unusual breadth across AI, UX, and backend systems.

Key Strengths

• Built AI interview/resume prep product end-to-end
• Startup accelerator recognition and prize winner
• Scaled student product to 200+ users
• Strong full-stack product ownership
• Applied AI/RAG implementation with privacy safeguards
• LLM output quality and hallucination control
• Compliance-aware product design in healthcare marketing
• Balances UX simplicity with legal/compliance clarity
• Performance optimization via caching and retrieval design
• Modular, maintainable coding practices
• Practical debugging and production issue resolution
• UI simplification and workflow-focused product polish
• Rapid execution through small iterative delivery
• End-to-end ownership of AI product features across frontend, backend, and database
• Strong React/TypeScript/Next.js product engineering
• Designs trustworthy AI UX for regulated domains
• Builds validation layers to improve LLM reliability
• Performance tuning across UI, API, and database layers
• Pragmatic product iteration under ambiguity
• Improves activation through guided user flows
• Creates reusable orchestration APIs from repeated workflow patterns
• Thoughtful multi-tenant architecture with safe configuration guardrails
• Debugs cross-layer integration issues with shared schemas and structured contracts
• Shipped end-to-end LLM agent (NextStep.AI) used by 150+ CUNY students
• Strong focus on reducing hallucinations via grounding/RAG and guardrails
• Designed multi-stage pipelines with stage-level metrics (latency, extraction success, fallback frequency, output quality)
• Improved performance and cost by switching to OCR-first extraction with LLM fallback
• Built deterministic workflows using schema-constrained outputs, validation checks, and human-in-the-loop review
• Implemented checkpointing/resume-from-last-state to handle intermittent failures
• Conservative automation for messy operational/AP-like data with normalization, confidence thresholds, and human review to prevent bad actions
• Database design for scalability/reliability (BCNF normalization)

Experience

Education

Skills

Languages