Experience

Professional Experience

2024 – Present

CirIAco

Lead Architect & Research Author

Independent Research • Cryptographic Knowledge Graph Engine
  • Architected a machine-native knowledge representation paradigm replacing human-centric documents with cryptographically verifiable atomic information units (Atmo blocks) in a Directed Acyclic Graph (DAG)
  • Designed a zero-trust provenance mechanism where factual propositions are SHA-256 hashed and parent-linked, eliminating reference drift and citation hallucination
  • Implemented recursive attribution traversal algorithms allowing inference engines to compute the exact informational lineage of multi-hop answers back to foundational truth nodes
  • Built deterministic graph traversal operators designed for fast context-window packing and low-overhead token consumption

2024 – Present

SumrAI

Core Systems & AI Infrastructure Engineer

Axtorch • Zero-Trust AI Gateway & FastMCP Runtime
  • Designed and implemented a heterogeneous dual-ring agent architecture separating autonomous ReAct reasoning loops (OpenFang/Rust) from sub-500ms token-streaming tool-execution pipelines (Django Ninja + Granian ASGI)
  • Engineered typed synchronous and asynchronous FastMCP servers enabling frontier models to safely execute parameterized operations, query vector databases, and invoke system tools across HTTP SSE and JSON-RPC
  • Formulated an information security architecture where sensitive entities are replaced with opaque cryptographic tokens (Rust FFI AES-256-GCM), preventing raw PII leakage to model context windows
  • Implemented hybrid retrieval pipelines over PostgreSQL 16 + pgvector, combining dense vector embeddings with metadata-filtered queries for context injection under strict token budgets

Jul 2024 – Present

NTT DATA

Systems Engineer – Enterprise Telemetry & Distributed Systems

Mission-Critical Financial Squads • BCP & Banco Pichincha
  • Managed 24/7 cross-squad telemetry and anomaly detection for high-volume banking payment rails (Client: BCP); monitored distributed service meshes using Dynatrace, Kibana, and Grafana
  • Profiled, tuned, and refactored enterprise Oracle and SQL Server query plans across millions of records (Client: Banco Pichincha), eliminating deadlock conditions and reducing query latency by up to 60%
  • Engineered custom Python integration bots for enterprise ITSM (ServiceNow API) over long-polling architectures, replacing manual workflows with reliable event-driven automation
  • Implemented defensive degradations, retry backoffs, and queue pacing during high-traffic transactional spikes across distributed banking integrations

2023 – 2024

Academic Research (UNPRG)

Principal Investigator & ML Engineer

Empirical Deep Learning & Signal Processing
  • Engineered an end-to-end computer vision pipeline using ResNet50 and ResUNet backbones, optimizing with custom Focal Tversky Loss to overcome severe class imbalance in brain MRI lesion segmentation; achieved Dice Similarity Coefficient >0.88 and IoU >0.80
  • Developed edge vision inference pipelines deploying quantized EfficientNet models on resource-constrained hardware (Raspberry Pi 4B), sustaining real-time processing (<200ms latency) without cloud backhaul
  • Conducted empirical PRISMA-compliant systematic reviews analyzing security and auditing frameworks across public cloud infrastructures (AWS/GCP)

Oct 2025

MAWJAC AI

Lead Infrastructure & ML Engineer

NASA Space Apps Challenge — Global Nominee
  • Recognized internationally as Global Nominee (top tier of 60,000+ global participants) for architectural design and execution of an AI-driven scientific visualization and telemetry prototype under intense hackathon constraints
  • Containerized ML inference pipelines with Docker, sustaining sub-200ms processing latencies for astronomical telemetry data

Current Research & Active Learning

Machine-Native Knowledge Representation

Deconstructing human document paradigms into cryptographically verifiable atomic blocks (Atmo) on DAGs to eliminate hallucination at the data structure level.

Model Context Protocol & Agentic Tooling

Engineering typed FastMCP runtimes for low-latency (<500ms) tool execution, pgvector hybrid retrieval, and deterministic multi-agent state handoffs.

Zero-Trust Vaults for Frontier LLMs

Formulating opaque entity pseudonymization via Rust FFI (PyO3) to ensure language models reason over relational semantics without ingesting raw PII.

Empirical ML & Hard Evaluation Frameworks

Designing stress tests and out-of-distribution evaluation benchmarks for vision and language models, moving beyond naive aggregate metrics.