Projects

Designed middleware between LLMs and tools so every tool invocation passes through authentication (JWT), role checks (RBAC), and rate limits instead of trusting the model with raw infrastructure access.

Runs tool handlers in a sandboxed execution model with structured token lifecycle management—revocation, rotation hooks, and audit-friendly boundaries suited for multi-tenant or regulated workloads.

Shipped a catalog of 8+ operational tools behind FastMCP Cloud patterns so agents can call APIs, databases, and utilities without exposing long-lived secrets or unconstrained network egress.

Built an end-to-end pipeline that ingests camera and sensor-derived signals, applies graph-style modeling (GNN) on behavioral relationships, and surfaces anomalies ranked by severity instead of naive threshold alarms.

Integrated vector search (Pinecone) with Gen-AI summarization so investigators get short, grounded narratives tied to similar past incidents rather than raw clip dumps.

Delivered multi-channel alerting (WhatsApp, email, phone) with escalation policies so on-call teams see concise context within seconds of a confirmed trigger.

Orchestrated ingestion of speech transcripts, on-screen OCR, and creator metadata so each video becomes a structured bundle LangGraph stages can reason over with deterministic fallbacks.

Grounded compliance reasoning with Azure AI Search and curated policy snippets—reducing “creative hallucination” when judging claims, disclaimers, or risky visuals.

Exposed reviewer dashboards via FastAPI + Next.js so legal and marketing teams can replay decisions, override labels, and export audit trails for regulators.

Automates the boring parts of Salesforce DX: scripted retrieve/deploy flows, repeatable org aliases, and guardrails before promoting bundles—so sandboxes, scratch orgs, and downstream CI jobs stay consistent.

Adds metadata-aware checks (diff-friendly summaries, destructive-change hints, and selective deploy targets) aimed at teams juggling Apex, Lightning Web Components, flows, and packaged dependencies.

Designed for pipelines: exit codes and structured logs plug into GitHub Actions or internal runners, turning “works on my machine” org tweaks into reviewable, reproducible steps.