● live
UnThink
AI-powered, fragment-first personal knowledge base and web capture engine.
Django 5.1FastAPIPostgreSQL 16Redis 7.2
all experience
● active[Apr 2026 – Present]
UnThink
Founder & Full-Stack Systems Engineer
Visit UnThink →- Founded and shipped UnThink end-to-end: an AI-native, fragment-first personal knowledge base SaaS with a published Manifest V3 Chrome Web Store extension and <300ms capture SLA.
- Architected a unified monorepo service architecture (Django 5.1 for ACID CRUD, quota enforcement, and auth; FastAPI for real-time Server-Sent Events streaming and async LLM routing).
- Engineered deterministic epoch versioning for multi-key Redis caching, invalidating complex multidimensional query caches logically in <1ms without Redis blocking or cache stampedes.
- Orchestrated 3 isolated Celery task queues for high-priority zero-shot AI classification, DOM parsing with Firecrawl fallback, and background knowledge digests with a multi-model failover cascade (Fireworks AI & Gemini 2.0 Flash).
- Hardened AWS EC2 infrastructure using non-root Docker containers, cap_drop: ALL, unprivileged Alpine Nginx gateway, and zero exposed internal ports with automated pre-deploy AST verification.
The story
idea → framing → build → deploy → what's next01 IDEA
Founded and shipped UnThink end-to-end
UnThink represents my work as a solo founder and full-stack systems engineer, building an AI-native second-brain platform from first principles. Traditional knowledge tools force developers to context-switch during deep work: saving a code snippet or insight requires selecting folders, writing titles, and assigning tags. The result is that over 85% of bookmarks are never opened again. UnThink inverts this: the atomic unit of capture is a fragment (2–8 lines of code or prose), and autonomous AI handles classification, tagging, and dual-directory filing at sub-300ms ingestion SLA.
02 FRAMING
The architecture was intentionally designed as a unified monorepo with specialized…
The architecture was intentionally designed as a unified monorepo with specialized service micro-orchestration. Django 5.1 and DRF manage transactional consistency, ACID CRUD, user authentication, tiered quota enforcement, and Stripe/Dodo payments. In parallel, FastAPI provides asynchronous, high-concurrency LLM streaming via Server-Sent Events (SSE). The services communicate through Redis pub/sub and three dedicated Celery queues (classification, scraping, and digest generation), ensuring transactional API responsiveness is never degraded by long-running AI inference.
03 BUILD
Solving cache invalidation was the primary scalability challenge in UnThink.
Solving cache invalidation was the primary scalability challenge in UnThink. With users querying paginated feeds, recursive folder trees with live fragment counts, and dynamic tag summaries, naive cache flushing caused severe Redis event loop contention and PostgreSQL stampedes. I implemented deterministic epoch versioning: bumping an atomic user-level version key logically invalidates all dependent query caches in sub-millisecond time without blocking Redis, while distributed mutex locks protect expensive tree aggregations during cache misses.
04 DEPLOY
On the client side, I engineered and published a Chromium Manifest V3 browser extension…
On the client side, I engineered and published a Chromium Manifest V3 browser extension to the Chrome Web Store. To handle MV3's 30-second ephemeral service worker lifecycle, authentication tokens and quota states are maintained statelessly via Chrome storage sync. A secure dashboard bridge synchronizes active sessions from letsunthink.tech with zero user friction, and the native zero-dependency popup loads in under 10ms.
05 WHAT'S NEXT
Outcome: an AI-native, fragment-first personal knowledge base SaaS with a pub…
Production deployment is containerized on AWS EC2 with an unprivileged Alpine Nginx reverse proxy. Every container runs under a non-root user profile with Linux kernel cap_drop: ALL enforced, with internal backend, AI, database, and Redis ports completely isolated from host exposure. Automated pre-deployment AST verification gates guarantee security invariants before any release reaches production.
Keywords
8 techs · links go to /stackRelated projects
1 project shipped alongside work at UnThinkRelated writing
11 posts reference UnThink- Linux Networking for Backend Engineers. Part 1: Namespaces and Virtual Interfaces
medium · 10 min read · linux · networking · docker · internals
- Linux Networking. Part 2: Communication From Namespace to Host Through Virtual Ethernet
medium · 9 min read · linux · networking · namespaces · veth
- Linux Networking. Part 3: Bridge and Default Gateway for One-Way Outbound Communication
medium · 8 min read · linux · networking · bridge · gateway
- Linux Networking. Part 4: SNAT and Packet Forwarding to Connect to the Internet From a Namespace
medium · 9 min read · linux · networking · nat · iptables
- PostgreSQL Part 1: Install, Setup, Creating Tables and CRUD Operations
medium · 7 min read · postgresql · database · sql · crud
- SQL Joins in PostgreSQL: Query Across Multiple Tables and Actually Understand What You Are Asking
medium · 8 min read · postgresql · sql · joins
- Aggregations & Grouping in SQL: The Queries Behind Every Dashboard and Report
medium · 8 min read · postgresql · sql · aggregations
- Your Easy Guide to Design Highly Available Systems With Redis. Part 1
medium · 8 min read · redis · high-availability · cache · distributed
- Step-by-Step Guide to Deploy Redis Cluster in AWS to Achieve High Availability. Part 2
medium · 10 min read · redis · aws · cluster · high-availability
- DrishtiAI: Building an AI Eye-Screening Agent for Rural India in 7 Days
medium · 10 min read · ai · computer-vision · webrtc · mobile
- Linux Networking for Backend Engineers. Part 1: Namespaces and Virtual Interfaces
native · 10 min read · linux · networking · docker · internals
Product case study
For the architecture deep-dive: three-layer build, isolation strategy, deployment topology. Read the case study: UnThink on /work.