Selected Work
The work, in outline.
Most of our engagements are governed by confidentiality. Below are representative archetypes — the shape of the work, the outcomes we hold ourselves to, and the deliverables that come with each. Detailed case studies, references, and code samples are available under NDA on request.
AI Systems
Production LLM platform for a B2B SaaS
Replaced an early-stage GPT-4 integration with a tiered routing layer (small model → frontier fallback), an embedding-based response cache, and a versioned eval harness. Quality held; cost and latency dropped.
Outcomes
−42%
LLM cost
−61%
p99 latency
+9pp
Eval pass-rate
Deliverables
- — Routing + caching layer with per-tenant budgets
- — Versioned eval set with CI regression checks
- — Observability: token, cost, and quality dashboards
- — Runbook + handover to client engineering team
Mobile
Flutter app for a regional financial services brand
Designed and shipped a cross-platform Flutter app from a clean slate: secure auth, biometric unlock, offline-first transactions with conflict-free sync, and platform-native scroll, share, and haptic affordances on both stores.
Outcomes
4.7+
App Store rating
<1.4s
Cold-start
99.7%
Crash-free sessions
Deliverables
- — Single Flutter codebase serving iOS and Android
- — Offline-first ledger with background sync
- — Fastlane CI/CD with TestFlight + Play internal channels
- — Post-launch crash and performance triage
Enterprise Web
Multi-tenant SaaS platform on Rails + Next.js
Architected and built a multi-tenant SaaS product from scratch: row-level isolated tenancy, role-based access, audit logging, billing integration, and a Next.js dashboard with edge-rendered marketing surfaces.
Outcomes
11 weeks
Time-to-first-customer
<150ms
p50 API response
100%
Zero-downtime deploys
Deliverables
- — Rails API with multi-tenant isolation and audit trails
- — Next.js dashboard + ISR-driven marketing pages
- — Stripe billing with metered usage and proration
- — SOC2-ready logging, access control, and data handling
AI + Web
Document-intelligence platform for a regulated industry
Built a hybrid retrieval pipeline (BM25 + dense embeddings + reranker) over millions of documents, surfaced behind a Next.js review interface with strict per-tenant isolation and full audit trails for regulator review.
Outcomes
+18pp
Top-1 retrieval
3.2×
Reviewer throughput
0
Audit findings
Deliverables
- — Hybrid retrieval pipeline with reranking
- — Per-tenant data isolation, encryption, and access logs
- — Next.js review and annotation UI
- — Regulator-ready audit and explainability surface
Want a closer look?
We're happy to share full case studies, code samples, and references under mutual NDA — just ask. Most discovery calls end with a tailored selection of relevant work.