AI Engineer specializing in RAG pipelines, LLM workflows, and full-stack SaaS development. Currently building real-world systems used in academic and product environments.
# context-aware retrieval over institutional knowledge
from langgraph.graph import StateGraph
from rag.retriever import HybridRetriever
graph = StateGraph(RAGState)
graph.add_node("retrieve", HybridRetriever(top_k=8))
graph.add_node("rerank", CrossEncoderReranker())
graph.add_node("answer", GroundedAnswerer(model="gpt-4o"))
# deterministic fallbacks → no hallucinated citations
app = graph.compile(checkpointer=PostgresSaver())
I work where AI, product engineering, and marketing systems overlap — building tools that move real metrics, not slide decks.
I design retrieval and reasoning systems that survive production — not demos.
End-to-end SaaS — from data model to deployed product.
Salesforce Marketing Cloud and n8n — beyond the standard activity catalogue.
Academic teams needed answers grounded in their own documents — syllabi, policies, archived research — without the hallucinations that off-the-shelf chatbots ship with.
Hybrid retriever (BM25 + dense embeddings) → cross-encoder rerank → grounded generation with citation enforcement. Deterministic fallback when confidence drops below threshold.
College project tracking was scattered across spreadsheets, email threads, and WhatsApp groups. Faculty couldn't see status; students couldn't see expectations.
Centralized workspace with AI-assisted progress summaries, milestone detection, and structured handoffs between students, mentors, and reviewers.
Marketing teams were losing high-intent leads to slow, manual triage. Enrichment, CRM, and outreach lived in three different tools — so first replies were always late and always generic.
Single n8n workflow handling webhook ingestion, AI-driven enrichment and qualification, HubSpot CRM sync, and LLM-generated personalized outreach — with retries, logging, and validation at every step.
Features are deliverables. Systems compound. Every component I ship should be reusable across two more contexts before it's done.
Code that another engineer can pick up at 2 a.m. always wins over a clever one-liner that requires me to explain it.
I use AI where it materially changes the outcome — not because the README mentions agents.
I work with founders and teams who need an engineer who can hold the full picture — from system design to the last line of production code.
See how I work →End-to-end design of LLM-powered features that ship to production users.
Drop reliable retrieval and reasoning into existing products without rewriting them.
React + Django product engineering — from idea to deployed v1.
Custom Journey Builder activities, AMPscript systems, and integrations.
Shipping BrainBench v1.2 with adaptive test sets.
AI-powered project workflows for ProjTrack Desk.
LLM orchestration patterns with LangGraph + state machines.
Whether you're hiring, exploring an AI product idea, or need a builder who can take it from zero to deployed — I'd like to hear about it.