Flagship platform
MPM Control Tower
Project lead & platform architect · Eye Care, TNT & Specialty · Pilot
An AI-enabled decision system for Medical Affairs operations that redesigns medical program management end to end. Tactical execution, spend and budget insight, travel governance monitoring, and financial investment planning and prioritization come together in one governed command center with a single knowledge source. The Travel Intelligence Hub and approval tracking sit alongside them, and a two-way Asana sync and the Smartsheet API keep plans and requests current.
Deterministic rules produce every number. the company's approved LLM gateway API, multi-agent orchestration and Graph RAG retrieval surface insights and risks on top of them, and an evaluator gate blocks any source-cited executive narrative that fails a fact check.
- ~170deterministic validation rules codify Medical Affairs standards
- 3 TAsEye Care, TNT and Specialty on one governed knowledge source
- ~1,000automated verification checks guard every release
- 1 gatestands between the agents and any executive narrative
- React
- TypeScript
- Vite
- Express
- LangGraph-style agents
- Graph RAG
- LLM gateway: GPT · Claude · Gemini
- Asana API (two-way)
- Smartsheet API
- Docker
- Corporate design system
The demo runs in your browser on invented sample data: no real data, people or systems are connected.
How it works
How a question becomes a decision
- Deterministic
- AI
- Gate
- Human
-
1 · Inputs
One knowledge source
Tactic plans and budgets, spend, travel requests and approvals, investment frameworks, field insights and strategy documents, with Asana and Smartsheet connected through their APIs.
-
2 · Deterministic validation
About 170 rules compute the numbers
Reconciliation checks, authoritative control totals and Medical Affairs standards, all calculated without an LLM.
-
3 · Retrieval
Knowledge graphs and Graph RAG
Multi-domain graphs with community detection give the agents grounded, filter-scoped context.
-
4 · Agentic AI
Multi-agent orchestration
A planner hands work to domain experts in parallel. A rule-based coverage check confirms nothing is missing before synthesis.
-
5 · Evaluator gate
Fact-checked before release
Every claim must cite its source. A failed check blocks the narrative.
Fails trigger one bounded retry
-
6 · Human decision
People decide and act
MPM and TA leads review the cited narrative and make the call. Every agent call is auditable.
Architecture
How data and AI move through the app
Everything runs in the analyst's browser: a React 18 and TypeScript app on the corporate design system, with every number computed client-side. A thin same-origin server hosts the bundle, proxies LLM gateway and Smartsheet, and mirrors data to SQLite. Six diagrams follow the data from upload to a defensible plan: select one to enlarge it, or open the interactive view with all six.
Choose a diagram
- IngestWorkbooks are parsed in the browser; tolerant header matching maps columns before anything is stored.
- PersistUploads land in IndexedDB and localStorage at once, then mirror to SQLite. Reloads read IndexedDB first.
- ComputeDeterministic engines turn the filtered data into KPIs, risk, gaps and reconciliation checks, with no LLM.
- GraphThe chat knowledge graph rebuilds when its data fingerprint changes; live Asana and Smartsheet rows join the build.
- Ask AIPrompts carry graph context and authoritative totals; the LLM client picks the LLM gateway route, and evaluators gate the output.
- Sync & exportFinancial Planning syncs the ranked plan to Asana (two-way) and pushes it to Smartsheet, with Excel, PDF and JSON exports.
Inside the platform
Twenty-four screens from the working build
Financial values, people's names, IDs and pipeline details are blurred. Select a screen to enlarge it.
Portfolio summaryControl totals for forecast, actuals, commitments and risk, broken down by sub-business unit.
Spend forecast by riskMonthly forecast stacked by risk level, with actuals to date and the current month marked.
Tactics by monthMonthly tactic spend stacked by risk, with a drill-down into one month's records, status and risk level.
Tactic groupsTactics grouped by spend category, with forecast, actuals, remaining, committed and planned totals and monthly totals per group.
Investment prioritizationCommitted and planned tactics ranked against strategy, with AI features that sync the ranked plan to Asana and Smartsheet.
AI Prioritize runA traced agent workflow: build the knowledge graph, validate in parallel, score alignment, write rationales, then evaluate and verify.
Financial planningThe 2027 Eye Care planning profile: targets against ranked committed and planned tactics, with priority type, status, category and gap links.
LRP alignmentLong-range-plan alignment by asset lifecycle archetype for US and International, with on-market mapping.
Tactic detailA tactic's strategic fit: mapped imperatives, asset fit, category and SI–SO validity scores, opened from the prioritization grid.
Investment knowledge graphFrameworks, assets, gaps and tactics linked in one graph, with community detection for Graph RAG.
Insights knowledge graphField insights become a live knowledge graph, with a chat panel grounded in the same graph.
Graph RAG chatConversational analysis across tactics, T&E, insights and financial data, with prompt libraries for risk, timing and stakeholders.
Maestro agentsTiered orchestration: planning, parallel experts, a coverage check, synthesis and a quality check that can send a draft back once.
T&E KPIsFilters and KPI cards for actuals, target, remaining and per-FTE spend, above the monthly trend with the highest-risk month flagged.
Travel Intelligence HubFrom request to spend in one view: travel requests grouped by approval stage, synced from Smartsheet, above monthly T&E by subfunction against target with the highest-risk month called out.
T&E by categorySpend by expense category, top travelers and spend-rate gauges by subfunction against budget.
Transaction distributionTransaction amounts by subfunction, with outliers and an AI summary of the distribution.
Spend patternsA parallel-coordinates view linking subfunction, traveler, spend and expense category.
Funding requestsIncremental funding requests with automated load validation and row-level flags.
Congress Hub timelineMilestones derived from the congress dates, with owners, phase tags and overdue tracking.
Gap trade-offsSeven-dimension scoring that places each gap in plan-and-resource, close-now or deprioritize.
Strategy map from AsanaAsana goals synced into a live strategy map, from company goal to team goals, sub-goals and milestones.
Settings & data sourcesData uploads with last-updated stamps per domain, an API connection test, and tabs for exports, the knowledge graph, LLM gateway, Asana and Smartsheet.
Modules I championed
Financial Planning & Investment Framework
Scores brand strategy, LRP alignment and Launch Guide readiness across 22 categories, surfacing over- and under-allocation before commitments lock.
Tactic Prioritization & Funding Rationale
Ranks every tactic on strategic fit, launch readiness and investment targets, with a defensible rationale per dollar and a flex list that lands spend on target.
Tactics Portfolio & Risk
Authoritative KPI control totals, automated reconciliation, uncontracted-charge watchlists and named accountability by business lead.
Travel Intelligence Hub
One view from request to spend. Travel requests submitted through a Smartsheet form land in a filterable tracker with international and domestic estimates, VP-approved totals and congress approval rates, rationale, approver comments and PDF or TSV export. Spend transparency across category, geography, vendor and compliance replaces periodic manual audits with attributed recommendations.
Knowledge Graph & Conversational Retrieval
Grounded, citation-enforced answers with filter-scoped authoritative totals.
Medical Insights & Care Gap Analytics
Insight-to-action analysis over field and DPO records, turning unstructured signal into prioritized care-gap and evidence decisions.
Gap Assessment (Anchor 6)
Validates and ranks evidence gaps on a seven-dimension impact-feasibility model.
Case studyCongress Journey & Expert Engagement Hub
Governed milestone scheduling, deliverables and expert engagement across the congress calendar.
Case studyApproval Process Management
Tracks a four-level approval chain (TA head, international head, BU VP and SPO) plus leader sign-off for congress travel, with approval history for funding requests and rules that flag missing decisions, amounts or justifications.
Asana & Smartsheet Integration
A two-way Asana sync pushes the ranked plan as tasks, custom fields, sections and AI summaries, and pulls status, budget and order back with field-by-field conflict review. The Smartsheet API reads requests and writes planning rows without duplicates.
LLM Insights & Risk Signals
the company's approved LLM gateway API through one client that switches models and retries. Prompts must quote the numbers the rules compute, and agents surface budget variance, uncontracted spend, T&E anomalies and pending approvals for people to act on.