Summer

Foundation: Sleep

Everything else runs on this. Fix sleep first.

Schedule: Bed 9:45pm, Sleep 10:30pm, Wake 7:00am, 8.5 hours

Why these times

Your body naturally wakes around 6:30–7:00am that’s your fixed biological anchor. Working backwards, 8.5 hours puts sleep onset at 10:30pm. You currently take ~45 minutes to fall asleep, so bed time is 9:45pm. As the routine kicks in over 2–3 weeks, this shifts naturally toward 10:00pm.

Going to bed before you’re drowsy is what built the can’t fall asleep problem. The bed needs to become a place your brain associates with sleep, not with lying awake.

Nightly schedule

The five fixes

1. Temperature Target 68–71°F (20°C). Fan circulating, not blowing directly at you. Keep feet outside the covers your feet and hands are your body’s main heat-release valves.

2. Brain dump (most underrated) Your brain treats open loops unfinished decisions, pending ideas as mild threats and stays alert to protect them. Writing everything down at 9:15pm tells your brain the information is safe. Write everything: ideas, worries, tomorrow’s tasks, decisions you haven’t made yet. Close the notebook. The effect on racing thoughts is significant.

3. Reading as cortisol-lowering Reading occupies just enough of your brain to stop it from generating its own thoughts no planning, no ruminating but it’s passive enough that drowsiness can build through it. Fiction works better than non-fiction because your brain can’t solve a story. After 1–2 weeks of consistency you’ll start getting drowsy mid-page.

Also: dim lighting from 9pm onwards. No music with lyrics or podcasts (your brain processes language like conversation keeps it alert). Slow breathing on high-stress days: 4-count inhale / 6-count exhale.

Avoid between 8:30–9:45pm: competitive games, planning, goal-setting, social media, exercise within 3 hours of bed.

4. Pre-bed snack Now that you’re in a bulk, use it. A small carb + protein snack 30–45 minutes before bed stabilizes blood glucose through the night and prevents the mid-sleep cortisol spike.

Good options: banana + peanut butter, rice cakes with honey, oats with milk, Greek yogurt with fruit.

5. Morning light anchor Get sunlight within 20–30 minutes of waking, every single day. Even 5 minutes outside. This resets your circadian clock and locks wake time at 7:00am consistently. It also makes falling asleep easier at night.

If you wake mid-night: don’t look at the time (turn phone face-down before bed). Stay still with eyes closed. If your mind races remind yourself the brain dump already captured everything.

Timeline

Pillar 1 Build

Ship to real users. No zero-distribution days.

Goal

Get something to real users and revenue. Not a perfect company a feedback loop.

You don’t need 14-hour days. You need consistency, focus, and compounding.

The product: Operational Momentum Infrastructure

Core insight: CRMs are passive memory systems. You’re building an active operational pressure system.

Companies don’t die because they lacked dashboards. They die because follow-ups died, deals stalled, timing slipped, nobody owned next steps, warm opportunities cooled off.

You’re not automating outreach. You’re automating momentum maintenance.

Positioning: Your GTM engine should not decay because humans got busy. / Never lose warm pipeline again.

Core job to be done: Every important conversation always has an owner, a next action, a deadline, escalation pressure, and momentum tracking.

ICP: 5–50 person founder-led B2B companies, agencies, recruiting firms, outbound-heavy SaaS, high-ticket sales teams. These teams already feel the pain, already spend money, and move fast.

The wedge: Prevent warm opportunities from dying. Simple. Painful. Measurable.

MVP core workflow

DO NOT build: multi-agent systems, complex orchestration, workflow builders, analytics platforms, custom CRMs.

You need ONE terrifyingly useful workflow. The emotional reaction: Holy shit, this recovered opportunities we would have lost.

Input sources: Gmail, Calendar, HubSpot/Pipedrive

What it does every day:

1. Scans conversations
2. Detects momentum decay
3. Identifies missing next actions
4. Scores opportunity risk
5. Drafts follow-ups
6. Revives stalled threads
7. Escalates overdue opportunities
8. Prioritizes who matters NOW

Core features:

Momentum Score every thread/account gets a state: warm / active / cooling / stale / dead / revived. Humans naturally understand momentum, not CRM stages.

Next Action Enforcement every meaningful conversation must have an owner, a next step, and a due date. If not, the product creates pressure.

Stale Opportunity Detection this lead replied 4 days ago, meeting happened but no follow-up, proposal sent but no continuation. This is the killer feature.

AI Follow-Up Drafts context-aware drafts based on the previous thread, relationship tone, stage, and timing. Short. Realistic. Non-AI sounding.

Revival Queue every morning: These opportunities are dying. This becomes the core habit loop. Not a dashboard. An action queue.

Distribution lead with proof:

• Your pipeline has 14 warm opportunities decaying right now.
• We found $42k of pipeline decay in 7 days.
• 32 warm leads had no owner.

That gets attention. Not AI SDR. Not better CRM.

Build order

Phase structure

May to June: Decide direction. Build MVP fast. Start outreach immediately. Optimize for conversations, not code perfection. No zero-distribution days.

June to July: Push distribution hard. Daily outreach/content. Get testers into calls. Try to hit first paying users.

July to August: Double down on what sticks. Kill weak features. Build onboarding. Turn best users into case studies.

Moat

Not AI. AI alone isn’t defensible. The moat becomes: operational graph, relationship state, momentum models, timing intelligence, behavioral data. Eventually you know when deals die, what follow-ups work, and what timing patterns look like. That’s defensible.

Pillar 2 Body

Gym daily. Meal prep on weekends. Build the system.

Goal

Lean bulk: ~0.5 lb/week. Better recovery, better sleep, better performance every month.

The biggest unlock is cooking autonomy stop depending on random food timing, family schedules, convenience food, and inconsistent protein.

The system

Weekend routine:

1. Grocery trip
2. Cook protein in bulk
3. Prep carbs
4. Portion containers for the week
5. Plan workouts

Weekdays: Low friction. No calorie guessing. Easy protein intake. Stable energy.

Core recipes to learn (start here)

You don’t need chef cooking. You need repeatable systems.

Learn 5–7 solid combinations that you can rotate without thinking. That alone is enough for an entire bulk.

Gym

Train hard. Recover hard. Then move on. Don’t let lifting consume the whole day mentally.

• Track performance monthly, not daily
• Daily movement is the priority consistency over intensity
• Sleep and food do more for your physique than extra sets

Pillar 3 Research

Context, memory, and continuity in AI systems. This is your long-term edge.

Research thesis

Build and study AI systems that maintain coherent long-term understanding over time.

Core question: How can AI systems sustain context, memory, and reasoning continuity without collapsing under long conversations, noisy retrieval, or irrelevant information?

The future of AI is probably not who has the biggest context window? it’s who manages memory, relevance, and continuity best? Intelligence across time is fundamentally a memory problem.

Main research themes

1. Memory systems

What should persist? What should decay? How should memories be ranked, evolved, and compressed over time?

• Working memory: current task, active goals, immediate conversation state
• Episodic memory: specific events (deployment failed yesterday, user changed direction)
• Semantic memory: abstracted long-term knowledge, user preferences, learned patterns
• Procedural memory: repeated task execution, successful strategies, learned processes

2. Context windows

Why large context windows alone are insufficient. Attention dilution, token inefficiency, context pollution, stale context. Can smaller but curated memory outperform huge context windows?

3. RAG systems

Not basic tutorials actual limitations and improvements. Why does naive RAG fail? How should temporal relevance work? How should conflicting memories be resolved?

• Basic RAG, then Temporal RAG (recency + importance + frequency weighting), then Hierarchical Retrieval, then Graph-Based Retrieval

4. Agentic systems

Reflection loops, planning loops, self-improvement loops, memory update loops. When should an agent reflect? How should planning evolve over time?

5. Forgetting and compression

A perfect memory system may be worse than selective memory. Humans forget intelligently.

Key concepts: memory half-life, salience scoring, relevance decay, abstraction layers, memory compression, memory reinforcement.

6. Long-horizon reasoning

Does the agent maintain continuity over days? Does it adapt correctly? Does it recognize state transitions? Does it hallucinate due to stale memory?

The summer project

Build two competing systems and compare them.

System A Stateless Agent Standard RAG, retrieval only, no evolving memory, no continuity model. Purpose: baseline.

System B Stateful Agent Persistent memory, temporal awareness, memory updating, summarization, importance scoring, context prioritization, memory decay, reflection loops. Purpose: test sustained intelligence.

Evaluation environment: Simulated dynamic worlds, not just benchmarks.

Example simulation software team:

• Day 1: backend issue discovered
• Day 3: deployment delayed
• Day 5: teammate unavailable
• Day 8: product requirements changed

Then evaluate: does the agent adapt? Does it remember unresolved problems? Does it use stale information?

Metrics

Quantitative: retrieval accuracy, hallucination rate, task completion rate, memory precision, context efficiency, token usage, latency

Qualitative: continuity, coherence, adaptability, perceived intelligence, reasoning consistency

Technical stack

• Models: OpenAI, Anthropic, open-source compare behavior
• Frameworks: LangChain, LangGraph, Vercel AI SDK, LlamaIndex
• Vector DBs: Pinecone, Weaviate, Chroma, Qdrant
• Storage: separate working memory / episodic memory / semantic memory

Key papers

Generative Agents, MemGPT, ReAct, Reflexion, Self-Refine, Voyager, LongMem, RET-LLM, Toolformer

Research notebook

Keep a living notebook not school-style notes. Instead: insights, hypotheses, architectures, diagrams, patterns, ideas for future systems. This may become one of your long-term unfair advantages. Most people treat AI as prompting. You’re already thinking about state continuity. That’s much deeper.

Deliverables by end of summer

Minimum: functioning stateful agent system, comparative experiments, memory architecture research, documented findings

Strong: open-source repo, research-style writeup, interactive demos, visualization dashboard

Exceptional: novel memory architecture ideas, publishable findings, recognition from AI communities

Pillar 4 Create

1 dedicated photography session per week. This is your creative reset.

Why this matters

Photography trains observation, taste, composition, patience, and noticing details. It balances all the time spent coding, optimizing, and thinking abstractly. Creative wandering is not optional it protects your mental state.

Philosophy

Most people only notice landmarks, dramatic moments, and obvious beauty.

This project focuses on: repetition, quietness, timing, routines, weather, movement, light, overlooked details, ordinary life.

The camera becomes a tool for slowing down and noticing. Finding extraordinary moments inside ordinary places.

Visual style

• Cinematic, documentary, restrained, minimal, atmospheric
• Composition: symmetry, leading lines, repetition, negative space, reflections, silhouettes, depth, geometry
• Best lighting: golden hour, rainy days, cloudy weather, blue hour, night lighting, fluorescent, fog, reflections after rain

Bad weather often creates stronger atmosphere. That’s an advantage.

Session structure

1. Explore bike or travel somewhere. No fixed expectations.
   • Train station, downtown, suburban roads, parking garage, bike path, café, strip mall, rainy neighborhood
2. Observe walk slowly. Don’t immediately shoot everything. Notice light, timing, reflections, movement, shadows, patterns.
3. Shoot take many photos. Most will fail. That’s normal.
4. Select after the session, choose only 3–10 strong images. Selection develops taste.
5. Edit slowly. Focus on mood, contrast, grain. Don’t over-edit. The image should feel real.
6. Archive store by date, location, session name. This becomes a visual memory system.

What to practice

Timing: Wait for movement, shadows, trains, people crossing frame, reflections changing, weather changes.

Light: Study golden hour, rainy reflections, harsh shadows, fluorescent lighting, window light.

Atmosphere: Try making photos feel calm, lonely, cinematic, reflective, quiet, transitional.

Storytelling: Create small visual sequences wide shot, detail shot, movement shot, reflection shot, closing image.

Caption style

Short, observational, restrained, honest. Avoid fake-deep writing.

Examples:

• The platform felt quieter after the rain.
• Most people ignored the reflection.
• The light stayed for maybe two minutes.
• Everything looked temporary here.

Album rules

• No forced epic photography. Ordinary places only.
• Emotion matters more than technical perfection.
• The environment should feel lived in.
• Capture moments most people walk past.
• Take photos because something feels interesting not because it looks trendy.

Daily Rhythm

Weekday structure

Weekly structure

Guiding Principles

Sleep is the foundation. Everything else compounds from it. Fix this before optimizing anything else.

No zero-distribution days. Even if you code 10 hours still post, still DM, still talk to users.

Protect deep work blocks. Even 2–4 focused hours/day is enough if done consistently. Most waste comes from context-switching, not lack of hours.

Don’t optimize for intensity every day. Optimize for sustained momentum for 3 months straight. That’s where compounding happens.

A successful summer is not: hitting huge revenue, publishing groundbreaking research, becoming massive physically, mastering everything.

A successful summer is: shipping consistently, improving physically, gaining independence through cooking and routine, deepening technical thinking, building stronger systems, coming out sharper than you entered.

That’s already a huge win.