Programs we've shipped.
Client engagements are described without naming clients — the specifics are real, the confidentiality is deliberate. References available on request.
Back-office platform for a NYC taxi-fleet operator
Situation. A venture-backed technology firm was building a full back-office platform for one of New York's largest yellow-cab fleet operators: medallions, vehicles, driver leases, weekly settlement, toll and violation processing, and payment integrations — live, with real driver pay and client funds at stake.
What HTS led
- Full program leadership: releases, sprint discipline, and daily prioritization across a distributed engineering team under production pressure
- Design of a correctness-first financial architecture — an immutable, append-only ledger with a single write path, reversing entries, SQL-view balances, and automated integrity self-checks
- Forensic diagnosis of financial-integrity defects: traced wrong-account and duplicate-balance bugs from raw data to a missing ledger-transfer primitive
- Executive-level client communication — including resolving a disputed claim with signed-document evidence and audit trails
Results
- Auditable financial architecture protecting driver pay and client money
- Production escalations handled with evidence-first triage — trust preserved through the hardest phase of the program
- A three-phase strategy (Stabilize → Extend → Transform) that reframed a single-client project into a productizable platform asset, with each phase structured as its own contract
A governed, multi-node agentic AI platform — designed, built, and operated
Situation. To prove the agent-directed delivery model end to end, HTS's principal designed and operates a production AI platform spanning three dedicated compute planes — control, data, and GPU inference — built almost entirely by directing AI agents rather than hand-coding.
What's running
- 40+ supervised services: agents, MCP servers, model gateways, watchdogs, and dashboards, managed 24/7
- 120+ MCP tools across four servers, making every platform capability agent-accessible
- Multi-model routing across frontier APIs and eight self-hosted models, with GPU vision inference
- Nine production databases plus a knowledge graph powering persistent, cross-agent memory
- Zero-trust edge: tunneled ingress, isolated VLANs, token allowlists, managed secrets
Why it matters to you
- 641 commits merged in 30 days by a single operator directing agents — through a gated CI pipeline (lint, tests, coverage, security scan, AI review)
- Every pattern we recommend to clients — agent governance, MCP architecture, evidence-gated delivery — runs here first, in production
- Full governance stack in daily use: gated CI (lint, tests, coverage, security scan, AI review), artifact-backed status, zero-trust access
The business case for agent-augmented delivery
Situation. A services firm's 12-person traditional agile team was slow and regression-prone. HTS built the case — grounded in measured results from its own platform — for restructuring to a 5–6 person senior, agent-augmented team.
The model
- Agents produce the majority of code and tests; senior engineers own architecture, judgment, and review
- A governed pipeline gates every merge: lint, tests, coverage, security scan, AI review — green merges, red iterates
- Trust built on external artifacts — commits, CI status, test evidence — not self-reported status
The target
- 3–5x delivery throughput at equal or better quality
- Smaller, more senior team with lower coordination overhead
- A delivery capability that becomes a margin and speed story for the business — not just an IT upgrade
Want the detailed version?
We're happy to walk through any of these engagements live — architecture, decisions, and what we'd do differently.