Conversational AI Services
Run conversational AI workflows with chatbots, voice AI, RAG-grounded answers, and lead qualification across WhatsApp, web, and internal systems.
Explore Conversational AIAI Operations Assistant
Persistent-Memory Operations Assistant
A construction business needed one operating layer across jobs, field photos, leads, schedules, and documents. We deployed a single-tenant OpenClaw assistant that worked through Telegram, preserved business memory, and drafted operational outputs without giving up approval control.
Hero Outcome
< 2hrs/day back
Estimated owner time reclaimed from faster job lookup, follow-up scanning, and first-draft document prep
System Snapshot
The visible product result was backed by a deliberately scoped architecture and delivery plan.
Interface
Telegram gave the owner a mobile-first way to use the system in trucks, on roofs, and between appointments without introducing another dashboard.
Telegram DMs
Voice Notes
Memory
The assistant keeps a structured workspace so context survives beyond one conversation and operational knowledge compounds instead of resetting.
SOUL / USER / AGENTS Files
Daily Memory Notes
Integrations
Existing business systems stayed in place. The assistant wrapped around them to assemble job context, schedules, photos, and lead activity when needed.
JobNimbus
CompanyCam
2 week delivery sprint
3 engineer build team
6+ production commits
Business Impact
Executive Outcomes
< 2hrs/day back
Less time rebuilding context across jobs, photos, schedules, leads, and documents
4draft workflows
Work orders, estimates, photo reports, and purchase orders no longer start from scratch
24/7assistant access
The owner can pull job context and next steps from the field instead of waiting to get back to a desk
100%approval gated
The system helps without sending or changing things on its own, which makes live adoption safer
The Challenge
“The owner was switching constantly between JobNimbus, CompanyCam, GoHighLevel, Google Workspace, and scattered conversations just to piece together what was happening across jobs. Important context lived inside tools and human memory instead of a reusable operating system.”
Job details, field photos, leads, schedules, and documents lived across separate tools, forcing the owner to rebuild context manually every day
Repeated instructions stayed trapped in calls, chats, and voice notes instead of becoming reusable business memory
Follow-ups and action items depended on personal memory, making it easy for important next steps to slip
Drafting work orders, estimates, purchase orders, and photo reports meant stitching information together by hand from multiple systems
Generic AI tools were too risky because they could not be trusted with sensitive folders or autonomous external writes
The Transformation
What changed after we built the system
Before
Job details, field photos, leads, schedules, and documents lived across separate tools, forcing the owner to rebuild context manually every day
After
One assistant can assemble job, photo, lead, schedule, and document context from the tools already running the business
Before
Repeated instructions stayed trapped in calls, chats, and voice notes instead of becoming reusable business memory
After
Persistent workspace memory and SOP capture turn repeated explanations into reusable operating knowledge over time
Before
Follow-ups and action items depended on personal memory, making it easy for important next steps to slip
After
Morning briefings and task scanning surface pending work before it disappears into chat history
Before
Drafting work orders, estimates, purchase orders, and photo reports meant stitching information together by hand from multiple systems
After
Operational documents now draft from connected system data through repeatable fabrication workflows instead of manual copy-paste
Before
Generic AI tools were too risky because they could not be trusted with sensitive folders or autonomous external writes
After
Approval-gated actions and a hard Google Drive exclusion make the assistant usable in a live business without giving up control
Why the memory layer mattered more than the model
The owner's real problem was not generating better answers to one-off prompts. It was carrying business context from one conversation, job, and follow-up into the next without rebuilding it every time.
A generic chatbot can help in the moment, but it does not become part of the operating system unless memory, routines, and reusable skills sit around it. That is why this build centered on persistent workspace files, daily memory, and SOP capture instead of only model selection.
Once the assistant became a memory layer, the value shifted from 'AI can draft text' to 'the business can keep adding context to the same operational system.' That is the part that compounds.
How We Built It
Technical architecture for the curious
Interface
Telegram gave the owner a mobile-first way to use the system in trucks, on roofs, and between appointments without introducing another dashboard.
Telegram DMsVoice NotesLong Polling
Memory
The assistant keeps a structured workspace so context survives beyond one conversation and operational knowledge compounds instead of resetting.
SOUL / USER / AGENTS FilesDaily Memory NotesSOP Builder
Integrations
Existing business systems stayed in place. The assistant wrapped around them to assemble job context, schedules, photos, and lead activity when needed.
JobNimbusCompanyCamGoHighLevelGoogle Drive + Calendar
Output
Real outputs mattered more than chat. The build generated operational PDFs and handled voice-note transcription so the assistant could create usable artifacts.
Typst PDF Generationwhisper.cppDraft Workflows
Operations
A lean deployment shape reduced maintenance risk while preserving upgrade, rollback, and handoff discipline through explicit scripts and runbooks.
Hetzner CX33systemd User ServiceProvisioning + Sync Scripts
OpenClaw
Codex
Telegram
JobNimbus
CompanyCam
GoHighLevel
Google Workspace
Typst
whisper.cpp
Hetzner
Bash
Python
Engineering Decisions
Tradeoffs we made and why
6commits
3engineer
2weeks
4test files
Single-tenant OpenClaw deployment on one Hetzner VPS instead of a heavier orchestrated stack
Benefit
Fast delivery, simpler handoff, and a supportable footprint for a small business owner
Cost
Less built-in multi-user scalability than a larger platform architecture
Telegram as the primary interface instead of a custom web app
Benefit
Matched how the owner already works in the field and removed adoption friction
Cost
Less structured UI control than a dedicated dashboard experience
Approval-gated external actions instead of autonomous writes
Benefit
Builds trust by keeping every outbound change visible and owner-controlled
Cost
Some workflows remain semi-automated because a human still approves the final step
Google permission boundary for the excluded Drive folder instead of prompt-only rules
Benefit
Sensitive content stays outside the assistant's reachable runtime path even if future prompts or skills change
Cost
More setup work and less flexibility when expanding folder access later
Document fabrication workflows instead of chat-only draft responses
Benefit
Produces field-ready outputs that teams can actually use downstream
Cost
Template maintenance is more involved than returning plain text in chat
OpenClaw is the underlying assistant framework. BrownMind handled the deployment architecture, workspace design, custom skills, integrations, security controls, and operational handoff for this client implementation.
Certain client names, proprietary workflows, screenshots, and internal assets referenced in this case study are protected under a non-disclosure agreement and have been anonymized or omitted to comply with our confidentiality obligations.
Need an AI assistant that fits real operations?
Book a free 30-minute call. We will map your current operating stack, find the biggest context-switching bottlenecks, and design an assistant that helps without creating new risk.
30 minutes with Apurva. Not a sales call.
Book Your Free Audit
Proof of Work
More Case Studies
AI Search Product
Business Impact
10+ hrs saved/week
Per team on information lookup Hybrid RAG Workflow Assistant
Gave teams trustworthy workflow answers in one chat interface instead of manual lookup across tools. Hybrid retrieval routes each query type to the right system for speed and accuracy.
Next.js Qdrant RAG AI SDK
Technical Proof
Before: Hours searching across tools
Now: Instant answers, one chat
Conversational Sales Automation
Business Impact
30% more revenue
From faster engagement and automated follow-ups Conversational Sales Automation
Turned inbound conversations into qualified opportunities with automated qualification, CRM updates, bookings, call handling, and follow-ups in one system.
Conversational AI Supabase Trigger.dev
Technical Proof
Before: Leads slipped away
Now: 30% more revenue
Financial Services Automation
Business Impact
3x faster onboarding
Clients go live in days instead of weeks Funding Onboarding Orchestrator
Removed cross-client data risk from a mission-critical onboarding workflow. Atomic isolation, dynamic field mapping, and secure PDF generation replaced brittle Zapier operations.
Trigger.dev GoHighLevel Google APIs
Technical Proof
Before: Zapier + data leaks
Now: Zero contamination