Proof, Not Slide Decks.

B2B SaaS company, ~$18M ARR, 12-person support team handling 400+ tickets/day across email, in-app chat, and a self-serve help center.

68% of tickets were repetitive tier-1 questions already answered in the knowledge base. Average resolution time had climbed to 14 hours and CSAT was declining quarter over quarter as the team drowned in volume.

A RAG-based support copilot integrated with Zendesk, the internal knowledge base, and Slack for escalations. Applying the Medha Protocol: Inputs (tickets, chat, KB articles), Intelligence (retrieval-augmented LLM routing), Memory (customer history and prior resolutions), Action (drafts replies, tags and routes tickets, triggers escalations), Oversight (human approval required on refunds and account changes).

Frontier LLM, vector database, Zendesk API, Slack, custom evaluation framework for reply accuracy.

Manager approval required on refunds and account changes. Full audit log of every AI-drafted and AI-sent reply. Weekly accuracy report reviewed with the support lead. Instant rollback to human-only routing if accuracy drops below threshold.

Extending the copilot to voice support and adding a proactive outreach agent that flags at-risk accounts before they file a ticket.

Commercial real estate firm, 200+ inbound leads/month across paid, referral, and organic channels, with a 2-person sales team.

80% of monthly leads were unqualified. Reps spent roughly 60% of their time on discovery calls that went nowhere, and hot prospects sometimes waited days for a first response.

A multi-agent qualification system: an enrichment agent pulls firmographic and intent data, a scoring agent ranks fit against the firm's ICP, a WhatsApp + email nurture agent handles warm-not-ready leads, and a scheduling agent books time directly with a rep. Hot prospects are handed off to a human within minutes.

Claude, HubSpot, enrichment APIs, WhatsApp Business API, Calendly, custom scoring model.

Every lead score and enrichment source is logged and auditable. Reps can override any score with one click. No agent contacts a lead outside firm-approved messaging templates without sign-off.

Adding a renewal/expansion agent that monitors existing tenant accounts for upsell and renewal signals.

Ecommerce operator running 3 Shopify stores with a shared 3PL and supplier network, processing several thousand orders per month.

Every order required 4 manual touchpoints between Shopify, inventory, the supplier portal, and the shipping carrier. Manual errors were driving a steady stream of refunds and support tickets.

An agentic workflow connecting Shopify, the inventory system, the supplier portal, and the shipping API — routing each order automatically while flagging out-of-stock, high-value, or COD orders for human approval before dispatch.

Multi-agent orchestration framework, Shopify API, internal inventory database, shipping aggregator, n8n.

Out-of-stock, high-value, and COD-flagged orders always route to a human for approval before dispatch. Full order-level audit trail from cart to carrier. Rollback to manual processing available per store in minutes.

Extending the same architecture to automate returns processing and supplier reorder triggers.