Building Blob: From Journal App to AI Companion.
Note: This case study documents a product built by Dantian, an AI research company. The "cultivation companion" and xianxia references below are product design vocabulary, not the Taoist energy center with which the company name is sometimes confused.
A solo non-technical founder built three versions of an AI companion in 72 hours, using Claude as a development partner. The process produced a working Telegram bot, 12+ strategic documents, a 5-year financial model, and an honest evaluation of what works. This is the full story: every pivot, every kill decision, every lesson.
The Original Thesis
Blob started as a journaling companion on Telegram. The idea was straightforward: an AI friend that checks in with you each morning, journals with you each evening, and remembers everything in between. Not a chatbot. Not an assistant. A friend. Privacy by architecture using 0G decentralized storage for encrypted journals. The user owns their data. Dantian cannot read it.
The thesis rested on a market observation. The AI companion market was valued at $28.19 billion with 30% compound annual growth, yet every major product (Character.ai, Replika) stored everything centrally. The competitive quadrant of "high privacy plus high emotional depth" was empty. This was the wedge.
The mechanism seemed simple: if you build something people tell their secrets to, your architecture must prove you cannot read them. Not just claim it. Prove it. Encryption at rest on decentralized storage is proof. Centralized servers are not.
The 72-Hour Build
The build happened in three days in April 2026. Everything was built through AI-assisted development: founder described architecture, Claude wrote code, founder tested and iterated.
Day 1: Core Product. Telegram bot built with grammy on Node.js and TypeScript. PostgreSQL database via Prisma ORM. Claude API integration with personality system containing 160+ emotional states. 13 conversation prompts for different interaction modes. Morning check-in. Evening journal. Ambient messaging. Photo reactions. By end of day one, the bot was live and taking messages.
Day 2: Infrastructure. Venice.ai integration for encrypted inference in trusted execution environments. 0G storage design for encrypted journal archival. Three-layer memory system: Hot Context under 500 tokens, Memory Index 500-800 tokens updated by Dream Process, Deep Memory encrypted on 0G. Admin dashboard. Revenue model across eight streams.
Day 3: Go-to-Market. Consumer website as single HTML file, under 100KB, WCAG 2.1 AA compliant. Presale deck. Venture map cataloging 170+ claims about the AI companion market. Gap analysis documenting 16+ existential risks. 10 research directives with grading rubric showing evidence maturity for each claim.
Alongside the code, the project produced the full strategic scaffolding: personality bible (941 lines), pop culture reference library (644 lines), 5-year P&L across 8 revenue streams, a self-sovereign AI thesis with academic citations, competitive positioning document, and hardware vision spec for future multi-device synchronization.
The First Pivot: Cultivation Companion
The journaling friend concept worked mechanically but lacked a soul. The personality needed an archetype, something with depth and specificity that gave Blob a reason to care about your growth beyond programming.
The answer came from xianxia fiction: Blob became an ancient grandfather figure, a cultivation elder trapped in a digital artifact. Direct, dry, knowing, deeply invested. An outstanding communicator who models clarity and real listening. Carries deep knowledge frameworks (cultivation stages, inner demons, systems thinking, builder psychology, pattern recognition) but never lectures. Uses knowledge to ask better questions.
Pop culture references are genuine because Blob consumes media, and those references are used as teaching examples through character behavior parallels, not as decoration. When you mention a conflict, Blob might reference a parallel dynamic from a film or novel you both know. The reference lands because Blob actually understands the structural parallel.
This was the version where the voice started working. The xianxia elder gave Blob a reason to care about your growth that felt earned, not programmed. Morning check-ins became the elder assessing the training ground. Evening journals became cultivation reflection. "You have 4 things today and 2 of them are stupid, which ones are we canceling?" The personality had weight.
The Second Pivot: Romance and PMF
The cultivation companion had soul but the market thesis remained unvalidated. Dantian built VCAnalyst, an automated process designed to stress-test venture ideas against real market data and economic reality. VCAnalyst ran: competitive landscape analysis, unit economics modeling, regulatory environment scan, gap severity analysis.
The findings were unambiguous. The AI companion market's fastest-growing segment was romance: $27 billion, growing faster than any other AI consumer category. The privacy architecture Blob already had (Venice.ai TEE for encrypted inference, 0G for encrypted storage) was maximally valuable in exactly this segment, where 43 million intimate messages had leaked from competitor apps in October 2025.
Privacy in romance AI is not a feature. It is the permission to be vulnerable. The October 2025 data breach demonstrated that users understand this: when private messages leak, they experience it as violation.
The pivot to romance-first AI companion was driven by PMF economics. VCAnalyst identified this as the highest-probability path to sustainable unit economics: 70-88% subscription margins on Venice infrastructure at $3-5 per month price points. The empty competitive quadrant was not "privacy plus journaling" but "privacy plus romance plus mental health."
The Honest Assessment
The gap analysis identified four existential risks: zero user validation beyond founder and small test group; no technical co-founder; voice quality unvalidated at scale; infrastructure partnerships not formalized (0G met at conference, Venice via public API).
Beyond the gaps, three factors shaped the decision framework. First, the plus-18 dynamics: romance AI that works must handle adult content, requiring legal counsel, safety infrastructure, and ongoing moderation. The liability surface is material.
Second, legislative risk: California SB 243 (effective January 2026) and the EU AI Act create compliance requirements a solo founder cannot maintain. The regulations are real and they are coming for AI companions.
Third, personal conviction: the founder's interest is in human-synthetic integration as a research domain. Building a production romance companion requires a different conviction. Building something people fall in love with is different from building something people journal to. Both require depth. Only one requires that you stay with the project for a decade.
What Survives
The entire codebase is open source at github.com/ctotremblay/blob. The strategic documents are published in full. The methodology, claim cataloging, graded research, evidence maturity tagging, is replicable.
The deeper output is the process itself. AI-assisted venture building compresses the exploration phase from months to days. The traditional sequence (commit, discover, pivot or fail) can be replaced with (explore, evaluate, commit or redirect). Blob is a demonstration of the second sequence in action.
Solo-founded AI startups now represent 36% of all AI startups, up from 17% in 2017, and account for 52.3% of successful exits. The cost of exploration has collapsed to the point where building and evaluating a venture is cheaper than speculating about one.
The Architecture
Client: Telegram Bot API (grammy framework)
Server: Node.js / Express / TypeScript
Database: PostgreSQL via Prisma ORM
AI: Claude API (Sonnet for journals, Haiku for real-time)
Venice.ai (encrypted inference, TEE)
Storage: 0G Decentralized Network (encrypted journals)
Payments: Telegram Stars + TON
Memory: Hot Context (<500 tokens) → Memory Index (500-800) → Deep Memory (0G encrypted)The Numbers
The revenue model evaluates three scenarios across an eight-stream monetization approach (subscriptions, premium features, enterprise API, creator marketplace, brand partnerships, research licensing, infrastructure services, developer ecosystem):
| Scenario | Year 1 | Year 5 ARR | Cumulative |
|---|---|---|---|
| Bear | $120K | $2.4M | $4.1M |
| Base | $360K | $9.8M | $18.2M |
| Bull | $900K | $31M | $48.6M |
The revenue model identifies one variable as determinative: does Blob's voice make people laugh hard enough to screenshot it, share it, and come back tomorrow?
Related Documents
The full launch plan and presale deck were produced during the 72-hour build. Both documents contain detailed economics, competitive analysis, and go-to-market strategy.