AI Engineer
Humanitarians AI · Mycroft Project
Developing an LLM inference gateway, benchmark harness, and evaluation tooling to examine quality, cost, and latency.
SIMBAREMUTEURO CHIN’OMBE
AI/ML Engineer · Generative AI & LLM Inference · Software Engineer
I build reliable AI systems, optimize inference performance, and turn complex workflows into useful software.

LLM systems / Inference / Agents
Developing an LLM inference gateway, benchmark harness, and evaluation tooling to examine quality, cost, and latency.
Built .NET Core services for HR and payroll. Optimized SQL queries and stored procedures, improving data retrieval latency by 15%.
Improved application performance by 20% through API and SQL optimization. Built Angular components, Redis caching, and secure authentication.
Developed capital markets applications, IPO workflows, stock-exchange integrations, and auditable financial data pipelines.
Delivered backend applications and third-party integrations for international clients using Python, .NET Core, and SQL.
Yeshiva University · New York, USA · 2024–2026
Harare Institute of Technology · Harare, Zimbabwe · 2015–2019
From GPU serving to agent workflows,
engineering beyond the model.
A containerized vLLM serving platform for Qwen2.5-3B-Instruct on an NVIDIA RTX 4080, deployed with Kubernetes and monitored with Prometheus and Grafana.
Benchmarked 1,812 requests across 43 configurations, varying prompt lengths and concurrency to measure time to first token, inter-token latency, throughput, and KV-cache capacity.
Added GPU resource allocation, health probes, and persistent model storage. Fixed container signal handling to reduce pod termination time from over 60 seconds to 1.35 seconds.
Peak aggregate throughput. Higher concurrency comes with per-user latency trade-offs.
An end-to-end LLM gateway with a six-task routing policy, three model tiers, output validation, and single-retry escalation.
Built as the sole engineer on the gateway, with per-attempt cost and latency logging and a frozen benchmark harness. Evaluation is underway against predefined cost and quality thresholds.
An autonomous Excel agent that turns requests into workbook operations through a chat interface, coordinating data and report agents.
A read-only SQL gate enforces single-SELECT validation, row caps, and timeouts. An Excel gate adds approval, overwrite protection, and an operation ledger for undo.
Typed plans, ground-truth checks, and an append-only event log support validation and replay. An evaluation harness uses a headless workbook simulator and deliberate failure injection.
Led a team in the Nx Video Solutions hackathon on HackerEarth, building an AI analytics system to measure crowd engagement and venue activity.
Processed a dataset of 20,000+ images to produce emotion, age, and crowd-density metrics. Used OpenCV and TensorFlow to support real-time video analysis.
Optimized the video pipeline with Nx Toolkit and delivered a dashboard to help event teams understand audience behavior.
Nx Video Solutions
HackerEarth hackathon
Applied across AI projects,
infrastructure, and software delivery.
Training, modeling, and applied computer vision.
Serving models and coordinating dependable AI workflows.
Deploying, observing, and benchmarking inference services.
Backend services, integrations, and interactive applications.
Working with transactional data and analytical pipelines.
Guardrails, automated checks, and controlled execution.
I’m Simbaremuteuro Chin’ombe, an AI/ML and software engineer based in New York, USA. My background spans capital markets applications, enterprise backends, and applied AI.
I earned my MS in Artificial Intelligence at Yeshiva University. Today, my work brings together model evaluation, inference infrastructure, and dependable agent systems.
Have an AI engineering role, a project, or a technical challenge in mind? Let’s talk.