One-page pre-class executive briefs for EMBA students — turning weekly work realities into actionable class prep and class learnings into actionable work leverage.

EMBA students are usually part-time students and full-time workers. Unlike traditional students, they're constantly asking: "How does this week's lesson help me make better decisions at work?" I built Delta Brief to create a positive feedback loop where weekly work realities sharpen in-class learning and discussion, and applied learnings from EMBA classes convert into tangible action items for the workplace.

Delta Brief generates a one-page pre-class executive brief for the next EMBA class session:

• Takes a 60-second weekly check-in ("what changed at work since last class?")
• Produces 3 ranked moves that matter, tied to the upcoming lesson's frameworks + learning objectives
• Personalizes via the student's role, constraints, and capstone goals
• Explicitly updates last week's decisions via: "Previously → Now → Update"
• Includes class discussion ammo and a concrete next action + deliverable

• Syllabus as source of truth: each session has topic, learning objectives, frameworks (names + bullets), prompts, and an assignment hook.
• MemMachine for persistent memory: We integrated MemMachine as our episodic memory backend to store and retrieve student context across sessions. Profile data (role, constraints, capstone) is stored as semantic memory, while weekly check-ins and generated briefs are stored as episodic memory anchored by session_id = class date. This enables true "compounding" where Week B can explicitly reference and update Week A's open threads.
• Strict brief template: forces structure and keeps it to one page.
• Compare view: Week A → Week B summary that highlights resolved threads and what changed.
• Guardrails: "no repeats" similarity checking, framework-name gating, and validation retries to keep the demo stable.

• Class-light outputs early on (generic advice): solved by enriching syllabus context
• Timeline confusion (brief dates drifting from syllabus): fixed by making session dates canonical
• Brittle parsing when relying on markdown: mitigated by storing structured fields (moves, highlights, thread updates) alongside the markdown
• Memory retrieval tuning: MemMachine's filter syntax required careful handling (single-quoted strings, correct response structure parsing) to reliably fetch prior context

• We made "compounding" visible, not just claimed: Week B clearly updates Week A with explicit thread resolution.
• Persistent memory that works: MemMachine stores weeks of context and enables the system to "remember" prior decisions, not just the last session.
• The system stays tight and usable: minimal input, one-page output, actionable next step.
• The demo has deterministic behavior through strict constraints (framework gating + no repeats + validation).

• Personalization only matters when it's obviously grounded in constraints and context the student cares about.
• Compounding requires explicit mechanics; without them, models "reset" every week.
• Episodic memory is powerful but needs careful structuring—anchoring memories to session dates made timeline reasoning reliable.

• Upgrade the memory layer to support longer-term use (semester-scale sessions, smarter retrieval, decay of stale context).
• Add a "ready for class" automated flow that generates each delta brief as a PDF artifact and sends it to the user's email inbox.
• Expand to multiple course scenarios and richer capstone deliverable support.
• Build an evaluation flywheel: track which brief recommendations led to action, and feed this signal back into MemMachine to improve future generation.

• Frontend: React, TypeScript, Vite, TailwindCSS, shadcn/ui
• LLM: OpenAI GPT-5-mini + text-embedding-3-large
• Memory: MemMachine (PostgreSQL + pgvector + Neo4j)
• Icons: Lucide React

1. Clone and install

   git clone <repo-url>
   cd delta-brief
   pnpm install

2. Configure Environment

   cp .env.example .env
   # Add your OpenAI API key to .env

3. Start MemMachine

   cd MemMachine-MemMachine-f85e49d
   cp .env.example .env
   # Edit .env and add your OPENAI_API_KEY
   ./memmachine-compose.sh

4. Configure app for MemMachine In your project root .env:

   VITE_MEMORY_PROVIDER=memmachine
   VITE_MEMMACHINE_BASE_URL=/memmachine
   VITE_MEMMACHINE_ORG_ID=patched
   VITE_MEMMACHINE_PROJECT_ID=delta-brief-demo
   

5. Run the app

   pnpm dev

┌─────────────────────────────────────────────────────────────┐
│                        Delta Brief                          │
├─────────────────────────────────────────────────────────────┤
│  React + Vite + TailwindCSS + shadcn/ui                     │
├─────────────────────────────────────────────────────────────┤
│  Memory Layer (MemMachine or LocalStorage)                  │
│  ├── Semantic Memory: Profile (role, constraints, capstone) │
│  └── Episodic Memory: Weekly check-ins, prior briefs        │
├─────────────────────────────────────────────────────────────┤
│  LLM (OpenAI GPT-5-mini) — Strict template + validation     │
└─────────────────────────────────────────────────────────────┘

MemMachine runs via Docker Compose with three services:

# View logs
docker compose -f MemMachine-MemMachine-f85e49d/docker-compose.yml logs -f

# Stop services
docker compose -f MemMachine-MemMachine-f85e49d/docker-compose.yml down

# Clean restart (wipes data)
docker compose -f MemMachine-MemMachine-f85e49d/docker-compose.yml down -v
./memmachine-compose.sh

1. Go to History and click Seed Demo Data.
2. Go to Brief.
3. Toggle Personalized (Memory ON).
4. Click Generate Brief.
5. Observe:
   • "Memory highlights" chips showing used context.
   • References to "Union pushback" (from Week A).
   • New moves based on "CTO resignation" (Week B delta).
6. Click Compare to see Week A → Week B side-by-side with thread resolution.

# Type check
pnpm check

# Lint
pnpm lint

# Build
pnpm build

MIT