Flagship client projectRetail & e-commerce
M&M Vinyl
A locally owned DTF printing and vinyl supply shop in Baton Rouge. We redesigned the storefront and built the owner a back office, AI-assisted from start to finish, and switched it all on with zero downtime.
The challenge
A same-day print shop can't go dark for a relaunch. The store's cart and checkout already worked, and customers already knew how to order a gang sheet. The owner wanted a new look and a clear view of the day without starting over or losing a single order.
What we built: the storefront
- A full redesign in a new black and mint brand look, built around the owner's official logo
- 165+ new pages: home, a shop, a DTF gang sheet page, resellers, teams and businesses, creators and DIY, visit, policies, and about
- A Learn section of how-to guides and calculators for decorators
- Real add to cart: products go straight into the shop's existing cart and checkout
- A catalog kept in sync with the live store, with prices and stock checked automatically and zero mismatches, and bulk pricing shown accurately
- A redesigned gang sheet page and a gang sheet builder restyled in the new brand, with the proven ordering flow left as it was
- A same-day cutoff clock that knows the shop's hours and holiday closures
What we built: the back office
- An owner dashboard with a daily board, orders, gang sheets, customers, and money in one place
- Secure text-message sign-in for the owner
- Shipping and policy wording made the same everywhere, with holiday hours and bulk pricing rules applied consistently
The launch
The new site went live with zero downtime and a one-click rollback ready. Before the switch, automated checks covered hundreds of pages and links, and a real test order ran all the way through checkout.
How we work
Silicon Bayou Designs designed and built both the storefront and the back office. The work was AI-assisted, which is how we build everything: AI handles the volume, like page builds, catalog checks, and testing, and we make the design and business calls and review every change before it ships.
- Stack
- Hand-coded HTML/CSS/JS, integration with the shop's existing online store, automated catalog checks, text-message sign-in
- Status
- Storefront and back office live as of November 2026. See it at mandmvinyl.com.
- A fit for
- Local shops with a working online store that want a new storefront and a better back office, without starting over or going offline.
Silicon Bayou productEnterprise compliance & advisory
AmonHen AI
An enterprise AI compliance and advisory platform, built for organizations that need certainty, not guesses. In production on iOS and the web.
The challenge
Compliance and advisory work means reading large volumes of documents against rules that differ from one domain to the next, and the answers have to be right. Generic chatbots guess; this work needs answers grounded in the documents and the rules.
What we built
- AI-powered document processing that ingests project documents
- A rule-based advisory engine with domain-specific knowledge packs and a knowledge graph engine
- Multi-domain compliance automation and real-time project intelligence
- A native iOS app (SwiftUI, MVVM architecture) with a project hub, AI suggestions, advisory chat, and quick actions
- A web app at app.amonhenai.com with an advisory interface, performance dashboard, and help documentation
- Secure authentication and enterprise-grade security throughout
- Stack
- Claude AI, FastAPI, Auth0, PostgreSQL, SwiftUI, vanilla JS web UI
- Status
- In production on iOS (App Store) and web. AmonHen also powers the Custom Chat Assistant add-on we sell with websites, including Bella, the AI host on Gino's Restaurant, and the chat on siliconbayoudesigns.com.
- A fit for
- Regulated or document-heavy organizations that need advisory answers grounded in their own documents and rules.
DemonstrationGovernment legal research
S.A.G.E. - Small Language Model for Legal Research
A legal research assistant for a state attorney general's office, built around a small language model we fine-tuned ourselves. Built and presented to the office's CIO.
The challenge
A legal office needs answers it can trace to the law. General-purpose chatbots invent citations, send questions to a third party, and write in their own voice instead of the office's. The office wanted to see what a small model, trained for its own work and hosted privately, could do.
What we built
- A small language model: Meta Llama 3.1 8B, 4-bit, fine-tuned with QLoRA so answers follow the office's structure, plain English for the reader and precise language for the record, with citations copied correctly
- Retrieval over public Georgia law (statutes, AG opinions, appellate opinions), chunked on section and subsection boundaries so every citation is pinpoint
- Hybrid search combining keyword (BM25) and vector retrieval, fused by rank, with a retrieval-confidence score and a plain "no authority found" instead of a guess
- Citation verification: anything cited that was not in the retrieved sources is stripped before the answer is returned
- A "learning" page that walks a non-technical audience through the training run: progress, loss curve, and a before and after comparison
- A research interface fronted by Sage, the office's owl assistant
The idea behind it
Fine-tuning teaches the model how to write, and retrieval supplies what is true. The model never has to memorize the law, which keeps it small, cheap to train, and easy to check.
- Stack
- Llama 3.1 8B, QLoRA, Hugging Face, ChromaDB (pgvector-ready), FastAPI, vanilla JS
- Status
- Working demonstration built for and presented to the CIO of a state attorney general's office. A pilot is the next step.
- A fit for
- Legal departments, government agencies, and compliance teams that want a private small model trained on their own document types.
Silicon Bayou productAccounting & CPA firms
Lodestar
A suite of six AI tools built for accounting firms, all live, from statement coding to return review.
The challenge
CPA firms spend staff hours on repeatable work: coding bank and card statements, answering IRS notices, onboarding clients, tracking filing deadlines, chasing missing documents, and reviewing returns.
What we built
- Keel, which codes bank and card statements to the right accounts, connected to QuickBooks Online
- A "Needs Your Judgment" view that puts items in front of an accountant for review
- A client view, firm-level secure access, and an audit log that records every action
- Dispatch, Charter, Watch, Manifest, and Sextant, covering IRS notices, client onboarding, filing deadlines, missing documents, and return review
- Stack
- Claude AI, QuickBooks Online, FastAPI, Auth0, AWS, SwiftUI
- Status
- All six tools are live: Keel, Dispatch, Charter, Watch, Manifest, and Sextant.
- A fit for
- Accounting, bookkeeping, and finance teams with high-volume, rules-based document work.
Silicon Bayou productConsumer health & fitness
M+ Health and Exercise
An AI fitness coach on the App Store: the fitness app that logs itself.
The challenge
Fitness and nutrition apps lose users to manual forms. Logging every workout, meal, and step by hand is the friction that makes people quit.
What we built
- Natural-language logging: type or speak "ran 3 miles, ate 6oz chicken" and AI parses it into steps, workouts, and meals
- A nutrition engine built on USDA data
- Adaptive training plans, a program library, a daily snapshot, and weekly trends
- HealthKit integration on iOS
- A marketing site, mplusfitness.com, that previews the programs and drives App Store downloads
- Stack
- Claude AI, HealthKit, FastAPI, Neon, Expo
- Status
- Live on the App Store. Marketing site live at mplusfitness.com.
- A fit for
- Consumer products where natural-language input can replace forms.
Client projectRestaurant & retail
Bernard's Specialty Meats
A Cajun specialty meat market in Alexandria, Louisiana, with online ordering and in-store menu boards running from one source of data.
The challenge
A retail counter with prices that change. Keeping the website, online ordering, and the menu boards in the store in sync by hand is slow and invites mistakes.
What we built
- A hand-coded website with the full menu
- Online ordering through Toast POS
- In-store digital TV menu boards driven by the same data, so a price change updates the website and the boards together
- Local SEO
- Stack
- Hand-coded HTML/CSS/JS, Toast POS integration
- Status
- Live at bernardsspecialtymeats.com.
- A fit for
- Restaurants and retailers on Toast or another POS who want one source of truth for prices across web and in-store.
Client projectRestaurant & hospitality
Gino's Restaurant
A Sicilian fine dining room on Bennington Avenue in Baton Rouge, run by the Marino family since 1966, with an AI host that answers guests around the clock.
The challenge
A long-established fine dining room needs a site that matches the room, and guests ask the same questions about the menu, wine, events, and private dining at all hours.
What we built
- A candlelit, old-world website with the full menu and the 300+ bottle wine cellar
- Thursday Jazz Night and a private dining inquiry form
- Bella, an AI host powered by AmonHen AI that answers guest questions around the clock
- Local SEO
- Stack
- Hand-coded HTML/CSS/JS, AmonHen AI (Bella chat host)
- Status
- Live at ginosrestaurant.com.
- A fit for
- Hospitality businesses that field repeat guest questions and want an AI host on their site.
Client projectCommunity & faith organizations
St. Aloysius Connect
A live parish engagement app for a Baton Rouge parish, on iOS and Android from one codebase.
The challenge
Parishioners need mass times, the weekly bulletin, the livestream, and announcements in one place on the phone they already carry, on both iPhone and Android.
What we built
- Mass times, weekly bulletins, and daily readings
- An in-app Vimeo livestream of mass
- Push notifications for announcements
- One cross-platform codebase shipped to both iOS and Android
- Stack
- React Native, Expo, Vimeo livestream, push notifications
- Status
- Live on the App Store, on iOS and Android.
- A fit for
- Churches, schools, clubs, and community organizations that need schedules, bulletins, livestreams, and announcements in an app.