
Suvidha is a conversational AI-powered shopping assistant designed to bring true convenience to online shopping and to empower users to make smarter shopping decisions — faster. Instead of relying on sponsored ads or generic reviews, Suvidha taps into authentic Reddit discussions, using a large language model to distill thousands of user experiences into concise, trustworthy insights. As users chat with Suvidha, it dynamically learns their preferences — like budget, brand loyalty, or feature priorities — and adapts its recommendations in real time. It also leverages smart caching to avoid redundant API calls, ensuring a seamless and responsive experience. Once a user has made an informed choice, the app offers a direct path to purchase by listing actionable product links — making the journey from confusion to confident checkout incredibly smooth. With Suvidha, we bring the power of real voices, personalization, and AI-driven speed together to deliver what online shopping should have always been: effortless, intelligent, and truly convenient.
8 Jul 2025

In today’s fast-paced world, we consume news in fragments — often just a headline or a single article. Pulse aims to bridge this gap by offering a smart, contextualised view of global events. Users can search or interact with recent happenings and receive real-time, LLM-powered insights beyond isolated articles. What makes Pulse unique is its local-first architecture, powered by RSS feeds, making it extendable to any domain where RSS is applicable — finance, health, or niche blogs. At its core is a fast incremental query engine (Feldora) that powers lightning-fast retrieval. Relevant articles are embedded and stored in Zilliz (Milvus), enabling semantic search. We built this as a prototype full-stack application using: • Rust server for performance-critical backend, • Python server for LLM orchestration (using Novita AI), • Next.js UI for intuitive interaction, • Traefik for reverse proxy and service routing, • TRAE for intelligent AI-assisted developer support across backend and frontend. Most importantly, we leveraged Trae's AI coding assistance to iterate rapidly across the stack — from auto-generating backend boilerplate, optimizing queries, to scaffolding frontend components. Trae helped us move faster and focus on logic rather than wiring.
15 Jun 2025