HaloITI

Spearheaded a B.Sc. Thesis R&D to modernize the university admissions system, designing and proposing a production-ready AI infrastructure for institutional deployment.

Engineered a zero-bloatware Autonomous RAG Agent in Pure Python (bypassing LangChain) with a Pydantic-based 4-Pillars Tooling ecosystem, enabling dynamic intent routing across Slot-Filling, Guardrails, Hybrid Vector Search, and native Text-to-SQL.

Slashed LLM latency and API costs by architecting a Double-Layer Global Cache combining a Pinecone Semantic Cache with dual-vector injection and a deterministic Tool-Level Redis Cache, secured by strict poison-prevention validation.

Containerized the entire architecture via Docker and deployed on a self-managed VPS, building a resilient infrastructure with a regex-driven Redis Circuit Breaker to intercept HTTP-429 rate limits and ensure 100% frontend uptime.

Eradicated context-window overflow and reduced token costs by >70% via a Redis-backed Sliding Window Memory algorithm, while asynchronously logging 100% of unbounded raw chat lineages into PostgreSQL for offline data audits.