AI infrastructure & software engineering, hands-on.

About

Engineers first, advisors second.

Tag Apps is a small team of independent senior consultants with more than ten years of hands-on work in artificial intelligence and machine learning. We’ve built predictive systems for energy, real estate, retail and finance; computer-vision platforms operating across global portfolios; and AI analytics for IT operations running at the scale of tens of thousands of devices.

Our work has always sat where machine learning meets real systems engineering — models that survive a Monday morning, vision pipelines that hold up across thousands of cameras, and trading and inference stacks that scale with traffic instead of with bills. We engage as a solo senior lead, embed a small pod inside your team for fixed sprints, or take an advisory seat alongside founders and CTOs.

• Independent. No resellers, no kickbacks — our recommendations are whatever is actually best for your stack.
• Senior team. Every consultant on the engagement brings 10+ years of production AI experience. No bench, no bait-and-switch.
• Production-minded. Every design choice is judged by whether it survives a 3am page, not whether it benchmarks well on Twitter.

Services

What we work on.

Engagements typically fall into one of these areas. Most projects mix two or three.

• AI cluster engineering

  Designing and operating multi-node GPU clusters — H100 / H200 / B200, A100, MI300 — with InfiniBand or RoCE fabrics, NCCL-tuned collectives, and Slurm or Kubernetes (Volcano, Kueue, KubeRay) scheduling for fair, preemptible access across teams.

• Distributed inference

  Production LLM serving on vLLM, TGI, SGLang and TensorRT-LLM with tensor / pipeline parallelism, paged KV-cache reuse, speculative decoding, and autoscaling tuned to real traffic shapes — not synthetic benchmarks.

• Storage & data pipelines

  Parallel filesystems (Lustre, WEKA, JuiceFS), high-throughput checkpointing, and dataset pipelines that keep six-figure-per-month GPUs saturated instead of idle on I/O.

• Cost & reliability

  Capacity planning across hyperscalers and neoclouds (Lambda, CoreWeave, Crusoe, Nebius), spot / on-demand mix, MTBF tracking on accelerators, and observability stacks (Prometheus, DCGM, Grafana, Loki) so failures are caught before a week-long training run is wasted.

• LLM applications

  End-to-end product builds — retrieval, evals, fine-tuning, agent orchestration, and the backend services around them. Built for teams that need to ship, not demos that need to impress.

• Advisory & due diligence

  Architecture reviews, technical due diligence for investors, hiring loops for ML/infra roles, and ongoing CTO-on-tap arrangements for early-stage teams.

Selected work

More than a decade of projects.

A snapshot of work the team has led across industries. Client names are kept confidential by default; specifics available on request under NDA.

1. 2024 — present IT operations · AI analytics

   AI analytics for large-scale IT deployments

   Predictive analytics platform for managed IT estates of 10,000+ devices, surfacing leading indicators of performance regressions, security anomalies, and usage-behavior drift across the fleet — turning telemetry that nobody reads into decisions operations teams can act on.

2. 2021 — 2024 Fintech · Trading

   AI-powered trading systems for fintechs

   Designed and operated ML-driven trading and execution systems for fintech clients — feature pipelines, model serving, and risk controls running on the kind of latency and uptime budget that doesn’t forgive shortcuts.

3. 2016 — 2020 Retail

   Predictive systems for retailers

   Demand forecasting, inventory optimization and customer-segmentation models deployed across multiple retail chains, integrated into the merchandising and supply-chain workflows that actually move the P&L.

4. 2015 — 2016 Computer vision

   Image-vision platform for one of the world’s largest mall operators

   Computer-vision pipelines across a global portfolio of shopping centers — foot-traffic analysis, anchor-store performance, and operational insights drawn from in-mall camera networks at scale.

5. 2014 — 2015 Real estate · Credit

   Predictive credit-scoring for a major LATAM real-estate group

   Credit-scoring engine used by one of Latin America’s largest real-estate companies to underwrite housing across emerging-market portfolios — replacing rule-of-thumb scoring with a calibrated, monitored ML pipeline.

6. 2010 — 2012 Energy · Oil & gas

   Advanced predictive systems for oil companies

   Some of the team’s earliest production work — predictive modeling for upstream oil operations, well before “AI” was a marketing term. The lessons from running models against messy, expensive, safety-critical data still inform how we ship today.

Stack

Tools we reach for.

Not exhaustive, and certainly not religious about any of it.