Computer Science + Mathematics Minor — University of South Florida
Agentic systems that reason over context, retrieve what matters, call real tools, and stop for a human before anything consequential.
Focus areas: Agent orchestration, Retrieval, Tool calling, Backend systems, Evaluation, Workflow design.
One interface. Several specialised systems. An orchestration layer deciding what happens next.
A local AI workspace where a single chat routes each request to the system that can actually answer it — a database, a document index, or an external service — and pauses for approval before it acts on your behalf.
Email the team our top accounts
Draft ready. Awaiting your send.
Retrieval and Calendar were available, considered, and not selected for this request.
Natural language goes in. A stateful execution pipeline decides how to query the database safely.
A question becomes a schema-aware query, runs against the database, and gets inspected before it is trusted. The pipeline checks its own work before it answers.
Question
Which regions grew fastest last quarter?
01 — Introspected schema
02 — Generated query
SELECT r.name, SUM(o.amount) AS revenue
FROM orders o
JOIN regions r USING (region_id)
WHERE o.quarter = ?
GROUP BY r.name
ORDER BY revenue DESC;
Parameterised · read-only · grounded in the introspected schema
03 — Returned rows
04 — Inspection before answering
An empty or malformed result routes back into the graph instead of forward into a confident sentence.
05 — Answer
Composed from the returned rows, with the executed query retained alongside it so the answer can be checked rather than trusted.
Retrieval is only half the problem. The other half is proving the answer was grounded.
A full RAG pipeline from document to answer, wrapped in a RAGAS evaluation harness — because a retrieval system you cannot score is a retrieval system you cannot improve.
What does the policy say about access?
Answered from the four retrieved chunks, each source kept alongside the answer.
Evaluated with RAGAS
Faithfulness
Supported by retrieved context
Answer Relevancy
Addresses the question asked
Context Precision
Retrieved chunks were useful
Context Recall
Nothing needed was missed
Different projects, one recurring shape: something has to decide what to do, do it, and check that it worked.
The routing logic that reads intent and decides which specialised system should handle it — and when to hand control back to a person.
Chunking, embedding, and vector search tuned so the model answers from grounded context instead of memory.
Typed, permissioned bridges between a language model and real services: databases, mail, calendars, repositories.
Graph-based execution where each node can inspect the run so far, validate its own output, and route accordingly.
Measurement around generated answers, because a retrieval system you cannot score is a retrieval system you cannot improve.
The underlying loop
2026
AI Intern
Worked alongside an engineering team on applied AI: LLM application patterns, retrieval over internal documents, and the practical gap between a prototype that answers questions and a system an organisation can rely on.
AI & LLM
Backend & Data
Interfaces
I am a Computer Science student at the University of South Florida, with a minor in Mathematics, focused on the engineering behind AI systems rather than the models themselves.
Most of my work sits in the same place: the layer between a language model and everything it needs to be useful — retrieval, tools, state, and the guardrails that decide when a person should be asked first.
I care about systems that are legible. If I cannot explain why the thing chose what it chose, I do not consider it finished.
Education
Open to internships and early-career roles in AI and software engineering.