Selected work / Engineering record

Ideas become
working systems

A closer look at the products, retrieval pipelines, and engineering experiments we build. Real projects, real trade-offs, and code you can inspect.

01 — Selected projects

A few things we've been building.

System / preview
▶
In development

REAL-TIME COLLABORATION

01

WatchTogether

Real-time YouTube watch parties. Rooms with 6-character codes, synced play/pause/seek, late joiners start at the correct position, and roles are enforced server-side.

The hard part

Cross-domain WebSocket authentication, server-authoritative playback time, and role checks on every event. Known limitations are documented in the README.

Under the hood

Next.js client with a Node/Express WebSocket server, PostgreSQL persistence, and JWT authentication. Playback state is broadcast from authoritative timestamps.

Next.jsTypeScriptWebSocketsNode.jsPostgreSQLJWT
System / preview
VECTOR
SPACE
✳
In development

RETRIEVAL INFRASTRUCTURE · OPEN SOURCE

02

RAGX

An experimental retrieval API for document ingestion, chunking, embeddings, pgvector similarity search, API keys, knowledge bases, and output guardrails.

The hard part

Connecting document ingestion, embedding generation, vector indexing, and output validation into one coherent pipeline. Hybrid retrieval and reranking are planned.

Under the hood

A modular TypeScript API managing chunks and embeddings in PostgreSQL with pgvector, API-key isolation, and server-side output evaluation filters.

TypeScriptEmbeddingspgvectorPostgreSQLLLMs
System / preview
In development

OPEN-SOURCE DEVELOPER PLATFORM

03

Devium

An open-source developer learning platform with devlogs and structured learning paths—built in public, one clear step at a time.

The hard part

Structuring modular Markdown learning paths and devlogs with clean navigation, readable content, and a consistent dark-mode experience.

Under the hood

Next.js App Router with static MDX content pipelines and responsive Tailwind design tokens.

Next.jsReactTypeScriptTailwind CSS
System / preview
⌕How do distributed caches work?
SOURCES FOUND · 03
In development

AI SEARCH & RETRIEVAL · OPEN SOURCE

04

Perplexity

A conversational search engine that retrieves current web information, evaluates sources, and synthesizes cited answers with streaming model responses.

The hard part

Building a parallel retrieval pipeline that parses search results, extracts useful context, evaluates relevance, and streams grounded citations.

Under the hood

A search and retrieval pipeline connected to a streaming LLM gateway, with web crawling, query reformulation, and server-sent events.

TypeScriptNext.jsLLMsSSEWeb Retrieval

Keep exploring

02 — Have a problem worth solving?

Let's turn the next idea into something real.

Have an MVP, a messy workflow, or a technical problem? Let's talk through the scope and find the simplest useful version to build.

Discuss your project