Where I build things that don't have a deadline.
Systems I've designed and shipped — from distributed architectures and TCP socket programming to generative AI models and AI education platforms. Some became products. All of them taught me something.
AI-Powered Indian Suit Pattern Design Generator
Built an end-to-end generative AI system for Indian female suit pattern design — fine-tuned an image generation model on domain-specific suit pattern data to produce culturally accurate, customisable design outputs.
Designed the full pipeline from user input (style, fabric, pattern preferences) to generated design output, with Stable Diffusion as the base model fine-tuned on Google Colab.
Near-deployment prototype built for real target users — independent suit makers and boutique designers.
Getting the fine-tuned model to produce culturally accurate patterns — Indian ethnic wear has subtle design conventions that generic models completely miss. Required careful dataset curation and iterative prompt engineering.
AI Study Notes & Learning Roadmap Generator · Jan 2026
Co-built and deployed a full-stack AI system generating structured study notes, learning roadmaps, curated resources, and Bloom's Taxonomy-aligned practice questions.
Fine-tuned LLaMA on Google Colab on a custom educational dataset for domain-specific accuracy. Chained LLM prompts with dynamic context injection, output parsing, and depth controls.
Presented to the Director General of NIELIT, MeitY (Government of India) — shortlisted for integration into government educational websites.
Designing the LLM chain architecture so outputs were consistently structured and pedagogically sound — not just plausible-sounding text, but genuinely useful study material aligned to Bloom's Taxonomy levels.
Concurrent Job Processing System
Designed and built a multi-worker distributed task queue from scratch — concurrent job execution across worker threads with a Redis-backed message broker for task persistence and dequeue ordering.
Implemented producer-consumer architecture with configurable worker pools, thread-safe task state management, retry logic with exponential backoff, and priority queue support using heap-based data structures.
Containerised with Docker Compose for reproducible multi-worker deployments; benchmarked throughput under concurrent load.
Identifying and resolving thread contention bottlenecks under high concurrent load — the interaction between Python's GIL, Redis round-trips, and worker thread synchronization created subtle performance cliffs.
TCP-Based In-Memory Key-Value Store
Built a Redis-inspired in-memory key-value store in pure Python using raw TCP socket programming — supporting GET, SET, DEL, EXPIRE, and TTL commands over a custom binary protocol.
Implemented thread-safe LRU eviction using OrderedDict + lock-based synchronization; added WAL (Write-Ahead Log) for crash recovery.
Benchmarked sub-millisecond latency under 500+ concurrent connections.
Making lock-based synchronization performant at 500+ concurrent connections — naive locking destroyed throughput, requiring careful lock granularity tuning and OrderedDict manipulation patterns.
Real-time Distributed Web Crawler with Worker Pools
Evolved WebScrapy into a distributed crawling system with a coordinator node managing URL queues via Redis and dispatching tasks to async worker pools — concurrent crawls across multiple domains simultaneously.
Built a real-time WebSocket dashboard streaming crawl progress and worker status live; implemented Redis Set deduplication, per-domain politeness delays, and structured PostgreSQL storage with normalised schema.
Voice-Activated AI Assistant
Built a wake-word-activated AI assistant integrating GPT-4, Wikipedia summarisation, OS-aware app launching, and a full TTS pipeline — async task handling for concurrent speech recognition and response generation.
Python, C/C++, JavaScript (Vanilla), SQL, HTML, CSS
Distributed Systems, Concurrency & Multi-threading, Async Programming (asyncio), TCP Socket Programming, LRU Cache, Producer-Consumer Patterns, WAL, DSA
LLM Fine-tuning (LLaMA), Stable Diffusion, Generative AI, LangChain, OpenAI API, AI Pipeline Design, Agentic AI
Google Colab, Google Opal, Google Flow
n8n, Shopify API, Redis, Docker, PostgreSQL, MySQL, Git, GitHub
Career Essentials in Generative AI — Microsoft (2024)