Selfology

Psychology coach gives accurate personalized answers because, like a regular one, it gradually gets to know you better

Faced immigration crisis in 2025 and wanted to create a coach for myself. Right now I can't afford therapists at €100 per session, and less qualified ones don't help. I took the Selfology questionnaire as a foundation and connected AI to it.

Before/After

Before: Therapy session once a week → $100/hour → expensive, or → $30/hour → generic advice like "live one day at a time" → feeling stuck.

After: Talk to AI anytime → $25/month → "I feel stuck" → AI finds your exact words from three months ago about fear and one month ago about perfectionism → connects dots you couldn't see. But most important for me — better understanding my desires and what exactly I should do, or what I truly want to do. AI remembers my answers and can object if I once again say "maybe this…". Unlike simple AI that gives generic advice like "go do it", here it's a coaching approach. Since I practiced coaching, for me Selfology is primarily a coaching project, though I also consider the psychological aspect.

Selfology helps me get through immigration crisis, midlife crisis at 35, and live more comfortably with ADHD and childhood trauma consequences.

Upgrade

Remembering my coaching experience and creating the Selfology questionnaire, I went further: I'm upgrading the onboarding. Now there are more questions — 1,513 questions across 17 life situations relevant in 2025, to quickly address what concerns you right now, something specific.

How It Works

Step 1. The system not only analyzes and remembers every conversation but also uses special onboarding based on the Selfology questionnaire, where I added 300+ more questions to cover all necessary aspects. Users go through onboarding with pleasure, at their own pace and volume. A special question router balances easy and difficult questions, continues themes if it sees interest. Claude analyzes answers and fills vector database with this analysis, forming a "digital personality fingerprint". AI learns how you think: analytically or emotionally, directly or delicately.

Step 2. Answer basic questions at your own pace or use one of 17 short quick-help programs. Not everything at once — AI clarifies what's important NOW, based on your state. Among programs: immigration stress, career change, "releasing baggage", FOMO, "AI is everywhere — what should I do?"

Step 3. Every answer is saved in memory. When months later you say "I'm stuck", AI searches your history, finds patterns across different conversations, and reflects them in new light.

Result. A coach who knows your history, adapts tone to your personality, and is available anytime. Asks better questions instead of giving generic advice.

Value & Scale

Solved for: one person — myself — during immigration crisis.

Here could be "market analytics", but I don't want to play that game: many companies try to build business on similar things, and results are weak so far. Theoretically the product could be worth millions, forecasts sound promising, but given the market state now, reality is far from bold predictions. This is not a certified psychotherapist, but a data and computation system.

Live product: Selfology v1 with questionnaire — 2021. Version v2 with AI memory launched in 2025 in debug mode.

Technical Architecture: How It's Built

1. Onboarding & Personality Mapping

• User starts with adapted Big Five test
• Original Selfology questionnaire forms onboarding foundation
• Questions presented gradually over days/weeks, not one heavy session
• AI analyzes response patterns and builds initial "personality vector"

2. Intelligent Question Routing

• System maintains state: what's answered, what's relevant now
• Router chooses next best question considering current emotional state, previous answers, and gaps in understanding
• No forced completion: some questions become relevant only after processing earlier material
• Can launch specific programs for specific situations

3. Vector Memory (RAG)

• Every answer converted to vector and stored in Qdrant
• Forms multidimensional representation of user psychology
• When new queries come, AI retrieves relevant context from entire history
• Example: "I'm stuck" → system finds conversation from three months ago about fear of change and one month ago about perfectionism → AI synthesizes personalized understanding

4. Adaptive AI Coaching

• Claude + GPT-4 combination: Claude for deep reasoning, GPT-4o-mini for conversational flow
• Personality profile influences communication style: direct or supportive, challenging or affirming
• AI doesn't give generic advice, but reflects your patterns in new framing
• Socratic method: fewer answers, more precise questions

5. Continuous Evolution

• "Personality vector" updates with every interaction
• System tracks changes over time: "how have my relationships with work changed over 6 months?"
• Long-term memory allows noticing shifts that aren't consciously registered

Key Decisions

• Curated questions instead of AI generation. Quality over quantity: 1,513 professionally crafted questions provide consistent depth versus scattered generated prompts
• Qdrant as vector DB. Conversation memory: AI "remembers" what you said 3 months ago
• Smart AI routing. 40% GPT-4o-mini, 50% GPT-4o, 10% Claude → ~$25 per user per month instead of $100+. Users can choose more expensive configuration if desired
• Persistent FSM on Redis. Progress isn't lost on bot restart. Big Five continues from where it stopped, not from beginning
• Workflows. Complex psychological analyses: tracking personality evolution, narrative generation

Real Numbers

Performance

• 1,513 curated questions across 17 programs
• Cost ~$25/month vs ~$100 per session
• 24/7 availability, 3-10 second responses
• 3 vector collections: personality, evolution, memory

What Actually Changed

• Before: therapists gave generic advice, I felt stuck
• After: AI remembers everything, sees patterns, asks questions adapted to my personality
• Can talk up to 3 hours a day instead of 1 hour per week
• Anonymity makes it easier to talk about difficult topics; yes, there are privacy concerns, but personally for me benefit is priority
• Actually helped get through immigration crisis when €30 therapy didn't help

What Makes It Special

Questions curated by hand, not AI-generated. 1,513 questions created for thoughtful reflection over days, not "quick run-through". Even without AI participation, this helps better self-discovery. Memory actually works: AI finds your answer from three months ago when today's struggle connects to it.

Skills Demonstrated

• Psychology: Big Five, CBT, ACT, Socratic method
• RAG system architecture with Qdrant
• Multi-AI orchestration: Claude + GPT-4 + GPT-4o-mini
• Content curation: 1,513 professional questions
• Production deployment: workflows, monitoring, GDPR
• Cost engineering: ~$25 per user through smart routing

Technical Stack

Technologies: Python, FastAPI, aiogram, Claude, GPT-4, GPT-4o-mini, Qdrant, PostgreSQL, Redis

Code: over 10,000 lines

Complexity: 10/10 for me. Written with Claude Code. Split into more than 32 subprojects.