This project is no longer active — this page is an archived record of the work done.
Archive · Jul 2023 – Jun 2024

An AI coach that actually gets how you think

Blissful AI started as a personal AI copilot for ADHD — empathetic, goal-driven, integrated into your calendar. Building it surfaced a fundamental gap in AI infrastructure: there was no good way to give LLMs persistent, contextual memory. We pivoted to solve that. This is a record of what we built, what we learned, and why we stopped.

Jul 2023 – Jun 2024 2 founders YC 2023 — Top 10% Archived

From personal pain to product

The original question was simple: why is it so hard to actually follow through on the things you care about? Not for lack of intent — for lack of a system that understands you, adapts to your energy, and keeps you moving without turning into yet another productivity app that demands the executive function it was supposed to help you build.

Sai and Shivanshu — a machine learning engineer and a product leader — had both lived this. They’d also watched friends and family struggle with ADHD in a world where the available tools were either pharmaceutical, expensive coaching, or to-do lists built for people who didn’t have the problem.

They interviewed approximately 35 people and 10 families across three segments — tech professionals with symptomatic ADHD, busy families, and university students — using structured usability tests and open-ended user research sessions. That research shaped every product decision: what the AI said first, how it framed goals, when it checked in. The coaching model drew on CBT frameworks, NIH-sponsored ADHD motivation research from Duke University, and Dr. Andrew Huberman’s work on goal-setting neuroscience.

The application reached the top 10% of Y Combinator applicants in 2023 — external validation of both the technical depth and the relevance of the problem.


The pivot: finding the real infrastructure problem

Building a personalized AI coaching agent forced a hard realization: LLMs are stateless, and the ecosystem had no good solution for it. Increasing context window size created latency and cost problems. RAG over interaction logs was a partial fix at best — it lacked human-like cognition and degraded over time. Topic summaries lost rich detail. Every approach had fundamental trade-offs, and there was no off-the-shelf memory layer to reach for.

This was 2023 — before MemGPT, before mem0, before the broader AI industry coalesced around “agent memory” as a distinct infrastructure problem. The team built their own solution and realized the problem was universal: nearly every AI application developer trying to build personalized, stateful agents was hitting the same wall. Pinecone’s sales team reached out independently, describing the exact gap their enterprise customers were trying to solve. This wasn’t a niche problem.

The pivot: build that memory infrastructure as a developer-first API. The closest legacy analogy was a CDP — but AI-native, covering short-term, long-term, episodic, factual, and procedural memory types through a single integration point.

~35
Users & families interviewed
2
Distinct products built
Top 10%
YC 2023 applicants

How it unfolded

Jul 2023

Idea & early exploration

Concept formed: AI assistant for busy professionals to manage goals, tasks, and life balance. Team of two, part-time, navigating visa constraints.

Sep – Oct 2023

Pivot to ADHD focus, prototype & user research

UI/UX mocks, click-through prototype, maze navigation tests. Structured interviews with ~35 people across tech professionals, busy families, and people with clinical and symptomatic ADHD. Research shaped coaching approach and product priorities.

Nov 2023 – Jan 2024

Full-stack product + waitlist

Built the complete system: FastAPI backend on GCP, Flutter mobile app (iOS/Android/web), Firebase auth and real-time data, OpenAI-powered coaching with function calling, Google Calendar integration, Pinecone RAG, stateful goal lifecycle management. Launched website and waitlist via word of mouth.

Early 2024

Insight: the memory problem is the product

Building a personalized AI agent surfaced a fundamental infrastructure gap that had no good solution — before this became a widely discussed problem in the AI ecosystem. Pinecone reached out independently with the same signal from enterprise customers. Team pivoted to Blissful AI Memory: a developer API for intelligent agent memory.

2024

Memory API + YC application — top 10%

Designed the Blissful AI Memory architecture covering multi-type memory (short-term, long-term, episodic, factual, procedural). Built POC. Applied to Y Combinator and ranked in the top 10% of applicants. The startup ultimately did not continue, but both the technical work and the market thesis proved well-founded.

What we were building

Phase 1

Personal AI Coach for ADHD

2023 – early 2024

A conversational AI coach built around the real challenges of ADHD: trouble initiating, time blindness, poor motivation regulation. The core problem with every existing productivity tool was that they required superhuman grit to use — the exact thing ADHD makes scarce. Blissful started from empathy before jumping to a plan.

The coaching loop: describe a goal in natural language → Blissful builds a structured plan (premise → vision → milestones → action items) → schedules it against your calendar with awareness of energy and commitments → checks in proactively with positive reinforcement, not guilt. Safety guardrails covered prompt-injection detection and sensitive-data filtering for the healthcare-adjacent use case.

Goal Lifecycle Management

Fuzzy goals converted into structured plans. Premise → Vision → Milestones → Action Items. Each stage stateful, resumable across sessions.

Empathy-First Coaching

Grounded in CBT and neuroscience. Goal-setting treated as a relational act — intrinsic motivation surfaced before tactics.

Calendar Intelligence

Google Calendar integration with priority-weighted scheduling across life dimensions: health, work, family, hobbies. Flexible and firm time blocks.

Positive Reinforcement Loop

Progress tracking across goal attainment and health metrics. Contextual motivation and accountability cues at the right moment, not on a fixed schedule.

Safety & Healthcare Guardrails

Prompt-injection detection, sensitive-data filtering, and curated mental health research corpora for responsible deployment in a wellbeing context.

Multi-Platform

iOS, Android, and web from a single Flutter codebase. Real-time sync via Firebase. Production deployment on GCP Cloud Run.

Phase 2 — Pivot

Blissful AI Memory API

2024

A developer-first API to give any AI application persistent, intelligent memory — making personalization possible without each team reinventing the infrastructure. The pitch: live in minutes, not months. No building or maintaining memory systems from scratch. The right tool for the job instead of RAG hacks.

Market context: This work predated the public emergence of tools like MemGPT and mem0. The team identified the gap from direct experience, then found independent validation from Pinecone (whose customers were actively requesting exactly this) and a projected $400B enterprise AI personalization market by 2028.

Multi-Type Memory

Short-term, long-term, episodic, factual, and procedural memory types — modeled on how human cognition actually works, not just vector lookup.

Simple API

One call to store context, one to retrieve. Intelligent prioritization and consolidation handled server-side. Async processing for low-latency responses.

Lower Cost than RAG

Reduces context window bloat and token costs for personalization use cases. Purpose-built vs. repurposed retrieval infrastructure.

User Control

Inspect and delete any stored memory via API. No opaque black boxes. Privacy-preserving by design, with SOC2/GDPR-ready vendor chain.

How the system fit together

The core engineering challenge was building a stateful, agentic LLM system on top of inherently stateless models. Each API call needed to reconstruct full conversational context, retrieve relevant long-term memory, determine the current stage of every active goal, and produce a response that felt continuous — all within latency constraints acceptable for a mobile app.

Client
Flutter iOS
Flutter Android
Flutter Web
API
FastAPI (Python)
Request Orchestrator
Goal Lifecycle Manager
Conversation Manager
AI Layer
OpenAI (function calling)
LangChain orchestration
Prompt engineering per goal stage
Memory
Embedding-based personal memory
Pinecone (vector DB)
Async consolidation
Firestore (goal state)
Integrations
Google Calendar API
Firebase Auth
GCP Pub/Sub
Secrets Manager
Infra
GCP Cloud Run
Terraform (IaC)
Docker
CircleCI

The goal lifecycle was the most complex piece: a single goal moved through stages (Premise → Vision → Milestones → Action Items → Scheduling) across multiple asynchronous conversations. Since each API call was stateless, the current stage was persisted to Firestore and read on every request to determine what the agent should do next. The memory layer stored behavioral traits, user preferences, and contextual facts as embeddings, with an async consolidation job that compressed older memories to control retrieval latency and token cost.


Backend

Python · FastAPI

RESTful API with async orchestration, LLM function calling, stateless goal lifecycle

Frontend

Flutter · Dart

Cross-platform iOS/Android/web. Real-time Firebase sync. Material Design

AI / LLM

OpenAI · LangChain

Conversational coaching with function calling. Custom prompt engineering per goal stage

Memory / RAG

Pinecone · Custom

Vector embeddings for knowledge base. Embedding-based personal memory with async consolidation

Data

Firestore · Firebase

Goal state, conversation history, user preferences. Firebase Auth and real-time sync

Infrastructure

GCP · Terraform

Cloud Run, Pub/Sub, Secrets Manager, Artifact Registry. Dev and prod as code

Integrations

Google Calendar API

OAuth-based access. Priority-weighted scheduling across life dimensions

CI/CD

CircleCI · Docker

Containerized deployment pipeline. Image pushed to GCP Artifact Registry on merge

What people said

Feedback gathered through structured usability testing and open-ended interviews with ~35 individuals and 10 families across tech professionals, busy families, people with clinical ADHD, and a Duke University clinician specializing in ADHD.

This is the best thing I have seen in a while that does something meaningfully helpful. I’m eagerly looking forward to getting on it.
Adam Christensen — Fintech Executive & Parent, Salt Lake City
There are so many busy founders who desperately need this today.
Kyle Maloney — Startup Co-Founder & CEO (YC W22), Berkeley CA
Whether you’re battling procrastination, living with ADHD, or not, it’s a must-have for a better, more productive life.
Cheryl Wu — Product Management Leader, San Francisco Bay Area
This looks like a promising product for adults with ADHD.
Professor & Clinician, Duke University Medical Center — Dept. of Psychiatry & Behavioral Sciences, ADHD Program

Who built it

Technical Co-Founder & CTO

Sai Jayakumar

LinkedIn →

~10 years of AI/ML experience across B2B, B2B2C, and Conversational AI. Led ML systems at Adobe. Architected the full backend and AI stack for Blissful end-to-end. MS in Data Science / Machine Learning from Carnegie Mellon University. B.E. in ECE from NIT Warangal, India.

Co-Founder & CEO

Shivanshu Singh

LinkedIn →

15+ years delivering data-driven products in fintech, martech, and healthtech. Head of Product at Marqeta. Led user research, product strategy, and go-to-market. PhD candidate in CS at Carnegie Mellon. MS in Software Engineering from SJSU.