Orbit

Challenge

A private capital operator wears a dozen hats — acquisitions, capital raising, lending relationships, property management oversight, legal, accounting — and each one carries its own web of active and dormant relationships. Without a system, follow-ups fall through the cracks, relationship context stays trapped inside email threads, vendor commitments silently disappear, and conference contacts go cold within 48 hours. Off-the-shelf CRMs solve this with more data-entry screens, which is exactly the work a one-person operation has no time for — the same lean-team problem Uplift Capital’s approach is built around.

Solution

Built Orbit — an AI-native relationship intelligence platform where the primary interface is conversational AI, not manual CRUD entry. It ingests Microsoft 365 data automatically and exposes it both to Claude (via an MCP server) and to a lightweight web dashboard.

• Microsoft Graph ingestion pipeline syncing email, calendar, and contacts via delegated OAuth (MSAL device-code flow), with incremental delta sync, automatic contact discovery from message headers, and smart filtering of no-reply/newsletter noise
• MCP server as the primary interface — Claude becomes the CRM, with typed tools for search, interaction history, follow-up suggestions, and email drafting; send-email is the only action gated behind explicit user confirmation
• Relationship health engine scoring each relationship on recency, frequency, reciprocity, response time, and engagement depth, with category-aware cadence expectations and longevity weighting
• “Orbital mechanics” modeling layer treating relationships as orbits — tracking health trajectory and velocity over time to flag ones drifting toward dormancy before they go cold
• Chief of Staff synthesis layer that combines commitments, open questions, thread state, and deal momentum into a prioritized next-best-action list and meeting talking points
• Daily briefing digest delivered by email, with a feedback loop (mute / pin / snooze) and a learning module that adapts categorization and cadence rules from user corrections over time
• FastAPI + React dashboard for at-a-glance contact, follow-up, and deal views, secured behind a private Tailscale network rather than a public auth surface

Impact

• Turned a manual-entry CRM problem into an ambient one — relationship data is captured automatically from the inbox, not typed in by hand
• Conversational querying means relationship context that used to take 5–10 minutes to reconstruct per person is now a single question to Claude
• Health scoring and the orbital decay model surface relationships going cold before they are lost, rather than after
• The Chief of Staff layer converts scattered signals — commitments, open threads, deal state — into a single prioritized action list each morning
• Human-in-the-loop by design: the system drafts and recommends, but every outbound email requires explicit confirmation before it sends