Production software, end-to-end. Deployed, not demoed.

What We Build

Core capabilities — production-grade, audit-first, AI-native.

Engagements

Pick the engagement that matches the lift.

Production AI agents. Custom-trained models. Cryptographic ledgers. Multi-tenant fintech. We’ve shipped all of it — in production, with paying users. We’ll ship the same thing for your business.

Industries

Where this work has landed.

A decade of software delivery across regulated and infrastructure-sensitive industries — the domains where correctness, auditability, and operational trust matter most.

Studio Products

What we ship for ourselves.

Hibiscus runs production systems on the same architecture we deliver to clients. Live, deployed, with paying users. Click through any card to see the architecture.

hyfstele · mlr-review

$review promo.pdf --vertical mlr

→ matched 3 OPDP enforcement precedents

overall_risk HIGH · 3 claims flagged

✓ signed: ml-dsa-65 + ed25519

✓ anchored: WORM block · 7-yr retention

→ every citation resolves to a real FDA letter

Defensible AI · Regulated · LIVE 2026

Hyfstele

AI you can put your name on. It flags, you decide, the record proves it.

Production AI that reviews high-stakes regulated work, grounds every answer in real enforcement precedent, and signs and anchors each decision into a tamper-evident record. Live at hyfstele.com.

See the architecture

• Four live verticals: pharma promotional / MLR review, manufacturing deviation / CAPA, aerospace nonconformance / NCR, defense in-transit visibility
• Grounds each answer in real published enforcement precedent (FDA OPDP letters, 483s, AS9100/FAA) and cites the specific source, refusing to assert anything it cannot resolve to a real document
• Every output cryptographically signed with a hybrid post-quantum + classical signature (ML-DSA-65 / FIPS 204 + Ed25519) and anchored to an append-only WORM ledger, independently verifiable offline years later
• Self-correcting and operator-in-the-loop: an independent model checks every verdict, and when the two disagree it flags the case for a human instead of guessing. It never acts autonomously on a high-stakes call.
• One agentic orchestrator routes by intent across verticals; onboarding a new domain is a corpus and an adapter, not new code

AI Agent · Real Estate · LIVE 2026

Sir Walter

An AI agent that runs your real-estate practice.

Drafts replies in your voice, nurtures your pipeline, coordinates your vendors. Live commercial subscription product at sirwalter.ai.

See the architecture

• Per-agent few-shot voice profile — every email he drafts is in YOUR voice, not a generic AI tone
• Vendor coordination in his own operator voice: terse, deadline-aware, no apology
• Lead nurture across SMS, email, and inbox with TCPA-compliant 9am–6pm timezone gating
• Drafts only — agent approves every send (NCREC-aligned, low liability footprint)
• Live at sirwalter.ai — founder annual $399/yr, monthly $49.99
• Built on the same Rust + TypeScript stack we deploy for clients

hiveworker · review.py

$python review.py --vertical mlr

→ loading owned model (sovereign)

→ retrieving precedent · RAG / pgvector

grader: 2 models agree · high confidence

✓ signed: ml-dsa-65 + ed25519

✓ anchored: WORM block

→ 3 low-confidence flagged for review

✓ HiveWorker - LIVE

Custom AI Model · 2026 · PRODUCTION

HiveWorker

We own the model. It goes toe to toe with the frontier.

A proprietary mix of owned, self-hosted models grounded with retrieval (RAG), matching the frontier API on our backtest while replacing rented model calls at a fraction of the cost. Self-hosted, your data never leaves. No vendor lock-in. The model layer most consultancies still draw on whiteboards.

See the architecture

• A proprietary mix of owned, self-hosted models grounded by a per-vertical RAG corpus, routed by intent so one runtime serves every domain
• A rigorous backtest harness that scores every output against historical regulatory actions, plus an independent grader that flags low-confidence verdicts for human review instead of guessing
• RAG over a per-vertical enforcement corpus (pgvector) so answers stay grounded in real precedent
• Replaces chained frontier-model API spend with low-cost dedicated inference
• Open base model, no vendor ToS concerns, no lock-in, fully ownable
• The same pattern we apply to client data: keep the data on your infrastructure, ship the model

Block #10242026-04-30

sha256:8a3f0c7e…c2e1

payment_received · $4,250.00

↓

Block #10252026-04-30

sha256:9b1c4d12…d3f2

vendor_payout · $1,400.00

↓

Block #1026WORM anchored ✓

sha256:a4d2871e…e7f9

contractor_complete · audit anchor

Cryptographic Audit Chain · 2025 · PRODUCTION

LedgerHive

Cryptographically verifiable proof of every action.

An append-only event-chain producing tamper-evident, independently verifiable proof-of-service records. The audit infrastructure for AI agents that touch real money. Externally anchorable.

See the architecture

• Append-only Merkle hash chain, modify one event and every subsequent hash breaks
• Blocks signed with a hybrid post-quantum + classical signature (ML-DSA-65 / FIPS 204 + Ed25519) and written once to a retention-locked WORM store, verifiable offline from the bytes and a public key alone
• No live lookup, no vendor dependency, the proof stands even if we disappear
• Currently anchors Hyfstele AI decisions, PayHive payment flows, and AI-agent action logs
• Chain-of-custody substrate for regulated industries: healthcare records, government data, financial settlement proofs
• The piece nobody else ships, and every regulated AI deployment will need within 36 months

9:41

☰ Messages

✉

Margaret Thompson

Follow-UpSMS

Post-service follow-up - weekly mowing completed this morning

✓ Send

✉

Rachel Foster

New LeadEMAIL

Inbound request via website form, wants bi-weekly mowing quote

✓ Send

PayHive

9:41

☰ Sir Walter

Ask Sir Walter

What should I focus on today? Who needs follow-up? Draft a reminder for overdue invoices What's my next opening? Any quotes about to expire?

Ask Sir Walter anything… ↑

PayHive

9:41

☰ Dashboard

Good morning, Blake.

7 jobs today. 6 messages ready to send.

Next: Thu, Apr 24 Week: 32 jobs

☔ Rain Thu & Fri. Sir Walter recommends front-loading - move 4 mowing jobs to Mon/Tue.

Crew status4/5 clocked in

Danny7:12 Marcus7:05 Tyler7:15

PayHive

Hive for Vendors · 2024 · LIVE

PayHive

The coordination + fintech platform for residential service companies.

Front-office, back-office, virtual assistant, and payments — what an HVAC, plumbing, landscape, or service trade runs their entire operation on. Live since 2024.

See the architecture

• Multi-tenant Stripe Connect platform — ACH, card processing, multi-tenant billing, payouts
• Front-office: scheduling, dispatch, vendor coordination, customer comms
• Back-office: invoicing, reconciliation, accounting integrations, ledger
• Virtual assistant layer: Sir Walter integrates as the AI agent for every customer-facing exchange
• Hive Suite plugs in: Listing Hive, Manage Hive, Hive CRM, Accounting Hive, Hive Sites — pick what fits your trade
• Production from first commit; regulated-environment-ready (PCI scope, KYC integrations, OFAC screening)
• Built specifically for residential service companies — the segment most AI tools ignore

Venture Studio · 2025 · LIVE

StudioHive

A venture studio that ships its own products.

From DSP Scorecard to Squadra HQ to Hive Worker, StudioHive ships and runs vertical-SaaS bets across delivery ops, athletics, voice AI, Firecracker sandboxes, and property management. We don't just consult on AI ops — we operate the whole portfolio.

See the portfolio

• Portfolio of vertical-SaaS bets — each shipped end-to-end (backend, frontend, payments, DNS, infrastructure, first paying customer) before deciding to scale
• Active products: Sir Walter (real estate AI), PayHive (residential services fintech), Squadra HQ (athletics-club registration), DSP Scorecard (delivery-service-partner ops)
• HiveWorker as the in-house custom AI model that powers it all
• Operating playbook: AI at the center, human approval gates, audit trail by default, no vendor lock-in
• Each product is a working reference implementation — what we'll productize for your business at the AI Agent Build tier
• The venture studio that built the things we'll ship for you

Claw · Operator Console

Pending approvals239

Active campaigns4

Today's revenue$3,420

Federal opps · high-fit7

OFAC screened100%

Autonomous Operations Runtime · 2025 · PRODUCTION

Claw

An autonomous AI agent watching every revenue stream, drafting every action, screening every send.

Claw is the production agent that runs StudioHive's revenue ops. Federal opportunity scanner, cash allocator, outreach orchestrator, OFAC screening — running today. The reference implementation behind every Build engagement.

See what it actually does

• Watches Stripe, listings, PostHog, and Thread Bank every 6 hours; ranks revenue opportunities by probability × estimated revenue using a frontier reasoning model
• Federal opportunity scanner: pulls SAM.gov solicitations, fits them against capability statement, auto-drafts Sources Sought responses for high-fit (≥0.55 fit score)
• Cash allocator: monitors incoming payments, splits revenue to operating, tax, and savings buckets via Thread Bank API or SMS approval
• Outreach orchestrator: drafts and dispatches campaigns via Instantly + Resend; hard-gates cold outreach to 9am–6pm prospect's local time
• Guardrails engine: every outbound action passes through OFAC SDN screening, FTC compliance checks, fair-housing classifier, and semantic rule matching before send
• 17-page admin console: dashboard, command (natural-language ops), opportunities, federal pipeline, allocations, OFAC review, replies, prospects, call list, settings
• Engagement invoice sender, product-to-Stripe sync, OFAC SDN refresher, RustLeads scraper + daily digest, send-queue dispatcher — all running today
• Currently 239 pending approvals queued for the operator — high-volume, low-latency, audit-first
• The reference implementation behind every AI Agent Build tier engagement

nectar · build

$nectar build app.nec --release

→ lex · parse · typecheck

→ IR optimization · WASM codegen

✓ runtime: ~3 KB gzipped

✓ tests: 2,248/2,248 passing

—————————————

vs Svelte 5 · 10K products:

sorts:   125× faster

filter:    5× faster

load:     3× faster

Programming Language · 2026 · OPEN SOURCE

Nectar

A web programming language we built. It compiles to WebAssembly.

A complete Rust-built compiler toolchain shipping a ~3 KB runtime, zero npm dependencies, and benchmarks 125× faster than Svelte 5 on production workloads. Open source. Live at buildnectar.com.

See the architecture

• Solo-built in a 4-week sprint (229 commits, March–April 2026) — full lex/parse/typecheck/IR/WASM-codegen pipeline in Rust
• Runtime: ~3 KB gzipped (vs React's 387 KB with 47 deps); a hand-written DOM syscall shim, zero npm packages
• Reactivity via O(1) signal updates and a command buffer — no virtual DOM tree-diff overhead
• Benchmarked side-by-side with Svelte 5 on a 10,000-product workload: 125× faster sorts, 5× faster filters, 3× faster initial load
• 2,248+ compiler tests passing — first-class type system, source-span error reporting, deterministic codegen
• Open source under github.com/HibiscusConsulting/nectar-lang
• Built to explore what a programming language for the AI-by-AI era looks like: type-safe, compile-time-verified, lightweight, portable to any WASM host

Federal Capability

Small-business-eligible across the codes that matter.

Hibiscus Consulting is small-business-eligible across relevant NAICS codes, with active SBIR submissions and a patent-pending cryptographic provenance primitive in autonomous systems, audit-provenance, and AI-agent deployment.