Tech Lead at Technergetics, working on the AI inference platform — Kubernetes, model serving, and the dev tooling around them.
- Retrieval policy — learns per-query search strategies, evaluated on BEIR
- HRM — recursive reasoning models for multi-agent pathfinding
- k8s homelab — Flux GitOps, Authelia, Grafana
Staff-level roles in AI infrastructure, platform, or MLOps. Remote-first, US-based.
Frank Cancedda
Tech Lead at Technergetics, where I work on the infrastructure ML teams ship on — Kubernetes, inference serving (vLLM, Triton), and the developer tooling around them. On the research side, I train small models for retrieval and reasoning. Off hours I run a self-hosted Kubernetes homelab and build Vivazo, a Rust game engine.
Compliance-grade dev environment, one command
Built the Tilt + k3d platform that brings the full 20+ service AI inference stack up locally, close enough to production that bugs surface in dev instead of staging. Team onboarding went from half a day to `tilt up`.
Hardened production AI inference for regulated workloads
Deployed and hardened NVIDIA Triton for inference in regulated environments. Designed end-to-end RAG pipelines: PGVector retrieval, TensorRT-LLM inference, JWT-secured Java Spring services.
GitLab CI/CD that cut deploy time and added auditability
Owned the release pipeline rebuild — GitLab CI/CD with build, test, scan, deploy stages. Cut deployment time, made every release auditable, removed the bus-factor from the original deploy scripts. Same pattern now powers the homelab GitOps.
I run my own k8s cluster.
A self-hosted Kubernetes cluster I built and operate end-to-end — GitOps deploys, real SSO, real observability, real CI/CD. It runs live applications, including a fitness tracker with AI meal analysis. It's how I keep the platform parts of my brain sharp on weekends.
Local k8s development platform (Tilt + k3d) mirroring production. One `tilt up` brings up 20+ microservices — vLLM, Triton, Keycloak, Istio, Minio, algorithm and prediction services.
A 1.6M-parameter policy network that learns per-query search strategies — query transformation and retrieval weighting — under a token budget. Built on my from-scratch PyTorch transformer lab; evaluated across four BEIR benchmarks.
Hierarchical Reasoning Model for Multi-Agent Path Finding, inspired by "Less is More: Recursive Reasoning with Tiny Networks." Per-agent metrics, heatmap traces, and interactive HTML visualizations.
Rust game engine (formerly OmniRx) with first-class WASM support and a formal API-stability/versioning system (Stable/Beta/Experimental tiers). Includes graph-based program-comprehension tooling.
- Triton Inference Server
- vLLM · TensorRT-LLM
- PyTorch · from-scratch transformers
- PGVector · OCR pipelines
- Kubernetes (EKS, k3d, OpenShift)
- Tilt · Flux · Helm · Terraform
- GitLab CI/CD
- Authelia / OIDC · Traefik · Istio
- Grafana · Prometheus
- Python — ML, services
- Go — control planes
- Rust — latency-critical paths
- TypeScript / Next.js — UI
- Java — enterprise services