Capabilities
The technical side of Honest Webbing, for the curious and for anyone vetting us.
If you're here to get a project priced, the estimate form is faster, and the work gallery is more useful. For very large projects, we have one of the best development companies on standby, ready to take your call, and we'll introduce you.
Honest Webbing is backed by 15+ years of building enterprise web applications across many industries: architecture and development in C#/.NET, Angular, Blazor, SQL Server, and Azure, plus the project side that makes software land — client communication, estimating, QA, and mentoring developers. Small-business sites get that same discipline, priced for a small business.
Custom business systems built on .NET / C#, Blazor, React, Angular, or Vue — with clean architecture your team can maintain after we're gone.
Hands-on experience moving MVC → Angular, WebForms → ASP.NET Core + Angular, and AngularJS → modern Angular. Same backend where possible, far less risk.
Experience backed by an Azure Developer Associate certification. Repos into Azure DevOps, CI/CD pipelines, Bicep infrastructure, Blob Storage, and failover that survives the bad day.
Local-first AI platforms: pgvector hybrid retrieval, NL-to-SQL, knowledge graphs, and local model serving on consumer hardware — no per-token bill shock.
The part most devs skip. Experience owning client engagement end-to-end — scoping, estimates, PM, QA oversight, delivery — across MVP through v3 releases.
4+ bootcamp cohorts trained in ABP Zero/IO, Angular, C#, and SQL. If you need a team that stays capable after the project, this is how you get one.
C# · TypeScript · JavaScript · Python · SQL · PowerShell
.NET / ASP.NET Core · Blazor Server + MudBlazor · ABP Framework (Zero/IO) · EF Core · Web API · SignalR · Background jobs
Angular 2–17 · React · Vue 2 · Bootstrap · PrimeNG · Metronic theming
SQL Server (DACPAC) · PostgreSQL + pgvector · Azure DevOps Pipelines · Bicep / IaaC · Blob Storage · Docker · CI/CD
RAG pipelines · Local LLM serving (llama.cpp/GGUF) · Embeddings & hybrid retrieval · NL-to-SQL · Knowledge graphs · MCP tooling
Supabase (Postgres + Auth + Storage) · Vercel deployment & edge functions · Bolt.new / Lovable-style app builders · Next.js App Router · Tailwind CSS v4 · Cursor / Copilot-assisted workflows · Resend transactional email
bUnit / Playwright E2E · Performance tuning · Client engagement & estimation · Project management · QA oversight · Team mentorship
Enterprise projects from the years before Honest Webbing was founded. Client names are withheld under NDA.
Architecture and hands-on delivery of a template-driven demand & purchase-order workflow platform with policy-based RBAC.
Full-stack ownership on an enterprise SaaS platform with tenant-scoped access and event-driven updates.
Python / PostgreSQL / pgvector hybrid-retrieval system: vector search + keyword search + query rewriting + NL-to-SQL + knowledge-graph traversal.
Windows-first serving platform running multiple GGUF models (1.5B–4B) on a single RTX 3050 4 GB via llama.cpp + CUDA.
Solo ownership across MVP → v3: client engagement, PM, QA, PTO, development. Sub-30-minute start-to-finish lending application delivered T&M.
Migrated a 7-year-old high-traffic portal while reusing the backend to accelerate delivery, then refactored to modern best practices.
We connect the tools you already pay for, and add AI only where it saves real time. A person checks anything before it reaches a customer. Our working loop is human architecture + machine throughput: people design the system and make the judgment calls, and AI compresses the hours between decisions.
A lot of “AI development” is marketing for cloud token bills. We run local-first AI infrastructure instead — open models on our own GPUs behind an OpenAI-compatible endpoint, wired straight into the editor. Client data stays on our machines during development, iteration is near-instant, and there's no per-token meter running while we build.
The same stack powers client-facing products: chat over your own documents with hybrid retrieval (vectors + keyword + query rewriting), plain-English questions answered from your database, and MCP endpoints so existing tools can talk to your data. When AI belongs in a product, we build it like plumbing — boring, testable, observable — not like a demo.
$ session --mode local
▸ architect: a person
▸ throughput: LLM-assisted
▸ models: GGUF @ localhost:8080
▸ data_egress: none during dev
▸ rag: pgvector + BM25 + rewrite
▸ nl_to_sql: yes
▸ mcp_endpoints: chat/analytics/health
# vram: 4GB. it works.
