Mittkin/Liquid AI Solutions

Custom AI agents, chatbots, and SaaS products Shipped end-to-end.

Production AI software built end-to-end — strategy, design, build, deploy, operate. We design conversational agents, configurators, scoring engines, and full SaaS products that compress hours of human work into seconds. Most builds ship in 2–6 weeks, on Vercel, with the right model (Claude, GPT, or both) wired into the tools that actually run your business.

What we build

From a chatbot on a marketing site to a full SaaS product, the surface changes — the stack doesn’t.

Conversational AI agents (chat, voice)
Configurators & cross-sell engines
Scoring & ranking systems
Document generation (proposals, RFQs, contracts)
Form & process automation
Marketing automation & lead scoring
SaaS products built from scratch
RAG / vector retrieval over your data
Live builds · 7 examples

Every card is a working production URL.

Click in, talk to the agent, configure a quote, generate a proposal. These are the actual products — not screenshots, not mockups.

SaaS product · Website · Marketing automation

GrantSkyNet

Full SaaS product built from scratch: AI agent that ingests Grants.gov, scores every open opportunity against an organization's capability profile, and drafts the application — executive summary, capability statement, prior performance, compliance matrix. Public marketing site, trial-flow drip campaigns, Stripe billing, Zoho CRM sync, multi-client management. Hours of grant-hunting reduced to a sorted shortlist.

SaaS productAI scoringDoc generationMarketing automationStripe + Zoho integration
grantskynet.comOpen
SaaS product · AI scoring · Compliance

GovconSkynet

Companion SaaS for federal contractors: processes SAM.gov in real time, scores 500k+ open contracts against a buyer's fit profile, and drafts capability statements, CO questions, and a compliance matrix on the SOW with amendment tracking. Same architectural pattern as GrantSkyNet, retargeted for the contracts side of federal work.

SaaS productAI compliance matrixReal-time data ingestionEmail send
govconskynet.comOpen
Conversational agent · Cross-sell · PDF quote

Current Tools — Product Specialist

AI chatbot for an electrical jobsite tool manufacturer. Customer describes the jobsite (cable size, conduit length, bend count); AI returns the right cable puller plus rope, sheaves, reel stand, and accessories — bundled into a branded PDF quote for distributor handoff.

ChatbotCross-sellPDF quoteDistributor handoff
current-tools.vercel.appOpen
Configurator · B2B sales handoff

Atlas Metal — Configuration Assistant

Configures custom serving lines for cafeterias and hospitals. Asks about facility type, daily volume, menu mix; returns specific model numbers, layout suggestions, and a sales-rep handoff with branded PDF.

ConfiguratorVisual specsSales-rep handoff
atlas-demo-peter-van-schaacks-projects.vercel.appOpen
Matching engine · Multi-stage workflows

Omni Search — Sourcing Assistant

Concept demo built on a Thomasnet/Xometry-style supplier directory. Buyer describes a part or asks an industry question; AI runs a short intake, then matches scored suppliers across processes, materials, certs, and geography — with multi-stage workflows for projects no single shop covers.

AI matchingMulti-stage workflowsBrowse + chat
sourcing-ai-demo.vercel.appOpen
Voice input · Live capture · Instant pricing

Xometry CNC — AI Sales Engineer

Concept demo built on Xometry's public CNC capabilities — not affiliated. Pre-qualifies CNC machining jobs before the buyer ever uploads CAD. Voice or text input, live spec-capture panel, instant ballpark price estimate, and routing to instant-quote or talk-to-sales.

Voice inputLive captureInstant pricing
xometry-cnc-demo.vercel.appOpen
Dual-mode chatbot · Sales + internal

RMC Solutions Guide / Field Copilot

Concept demo for Rochester Midland Corporation, a specialty chemicals company. Single AI engine, two audiences: website visitors get a friendly Solutions Guide that captures leads for site assessments; internal RMC field reps get a senior-engineer copilot with chemistry recommendations, troubleshooting flows, and compliance guidance.

Dual-modeVoice inputLive captureLead handoff
rmc-field-copilot.vercel.appOpen
How an engagement runs

Scope, prototype, ship, operate. Usually 2–6 weeks to first production deploy.

01
Scope

A 30-minute call to pin down the friction, the data sources, and what 'shipped' actually looks like for your team.

02
Prototype

A working prototype in week one or two. Real model, real data, real interaction — no slides.

03
Ship

Deployed to Vercel under your domain, wired to your CRM, email, and billing. You own everything.

04
Operate

We monitor, tune, and extend. AI products drift; we keep yours sharp as models, data, and your business change.

The stack

Modern, boring, production-tested.

Models
  • Claude Sonnet & Haiku
  • GPT-4 / 5
  • Voice (Whisper, ElevenLabs)
Build & deploy
  • Next.js + React + TypeScript
  • Vercel (edge + functions)
  • Supabase / Postgres
Auth & billing
  • Clerk
  • Stripe
  • Custom RBAC
Data & search
  • RAG / pgvector
  • Structured tool use
  • Real-time ingestion
Integrations
  • Zoho CRM / Books / Campaigns
  • Gmail, Calendar, Drive
  • SAM.gov, Grants.gov, custom APIs
Output
  • PDF / Word / PPTX generation
  • Email send + drip
  • Distributor / sales handoff
Where we’ve worked

Patterns repeat across industries. The wiring is the same.

Industrial manufacturing
Specialty chemicals
Federal contractors
Non-profits & research orgs
B2B distributors
Professional services
Pair with Liquid CRM

Running Zoho One? We can wire the AI directly into it.

Lead scoring, follow-up drafts, deal forecasting — the same studio handles both halves.

Liquid CRM Solutions
Start a build

One conversation. Working prototype in two weeks.

Tell us where the friction is — a quote process, a search, a customer-support backlog, a SaaS product you want to ship. We’ll come back with scope and timeline within a day.