LLM Fusion Study
Empirical study comparing fused multi-model panels with judge aggregation against single frontier models. Python harness with OpenRouter integration, standard benchmarks (DRACO, GSM8K, GPQA, HumanEval), and run-level cost tracking.
Product thinking across web, mobile, and AI.
Empirical study comparing fused multi-model panels with judge aggregation against single frontier models. Python harness with OpenRouter integration, standard benchmarks (DRACO, GSM8K, GPQA, HumanEval), and run-level cost tracking.
Backend API for event photo discovery. Face detection and embedding with dlib, identity clustering via streaming centroids, pgvector cosine search, and selfie-based authentication. Deployed with Docker on Render.
Personal knowledge system with a custom MCP server, FTS5 full-text search, a tag and rules engine, and a PARA-organized markdown vault backed by SQLite.
Next.js landing page connected to MCP tooling. Prototype for exposing agent-accessible tools and context through a web interface.
MCP server implementation in Go. Connects LLM clients to external tools and data sources over the Model Context Protocol.
Expo mobile application with Groq-backed chat, Google Calendar and Gmail integration, Meet scheduling, voice input, and daily briefings.
React Native journaling app with OpenAI Whisper transcription, sentiment analysis, weekly summaries, and streak tracking. Node.js and PostgreSQL backend.
Next.js storefront with infinite scroll, ISR caching, JSON-LD structured data, NextAuth authentication, Supabase backend, and Zustand state management.
Anonymous campus rumor verification platform with Bayesian trust scoring, cryptographic voting, evidence uploads, and Gemini-based content moderation.
Computer science student at NUST. I care about shaping useful product ideas, validating them quickly, and building enough of the system to make the experience real.
Recent work includes an empirical study of LLM fusion on OpenRouter, MCP servers for agent tooling, and a FastAPI service that uses pgvector for face-based photo retrieval at scale.
I like working at the intersection of product, design, and technical execution, especially where AI can make workflows simpler or more personal.
LLM fusion research, MCP server development, and vector search backends with FastAPI and pgvector.
Web and mobile applications including e-commerce, integrated assistants, and hackathon projects. Next.js, Supabase, and React Native.