Hi, I'm Ahsen

Research without shipping is a paper. Shipping without research is noise. I do both.

The gap between research and production is where most people live. I close it.

View Work ↓Book a Call →

Co-founder · Velton

AI Research Intern · UNB

AI Engineer · CoralOS

Projects

IntelliComm ProtocolIn Progress

Jan 2026

A pip-installable, agent-agnostic Python framework that turns any LLM agent into a monetized, pay-per-call endpoint. Dual-protocol (Socket + HTTP/REST) with lightweight decorators, x402 micropayments, and zero manual transaction logic across Base, Ethereum, and Solana.

Python,FastAPI,x402,eth-account,Uvicorn,Rich

Trend-X-BTC

Oct 2024

30-day Bitcoin forecasting SaaS — LSTM + Multi-Head Attention trained on market sentiment, on-chain activity, macro factors, and trading volumes ingested from Binance, CryptoCompare, and Google BigQuery. Daily-refreshed predictions piped to Firebase; LLM layer (LLaMA, Gemini, Mistral) overlays real-time market context.

PyTorch,LSTM + Attention,Next.js,Firebase,Binance API,BigQuery

Eeko AI

Sep 2024

AI-powered agriculture assistant for smallholder farmers. Llama 3.2 Vision for pest and disease detection, YOLOv5 for real-time weed identification, and a contextual LLM over the NASA LARC satellite API. English-to-Urdu speech model bolted on for accessibility. Reached 439+ visitors within hours of launch.

Llama 3.2 Vision,YOLOv5,NASA LARC,Next.js,Hugging Face

VisionPlay

Oct 2024

Real-time football match analysis from broadcast footage. Fine-tuned YOLO + ByteTrack for player, goalkeeper, and ball detection. K-means for team assignment. Optical flow corrects camera motion before mapping pixel coords to field positions. Live possession, speed, and distance per player.

YOLO,ByteTrack,OpenCV,K-means,Python

Research

2026 · PreprintCoralOS Research

Beyond Rule-Based Workflows: An Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent Communication

Xinxing Ren, Quagmire Zang, Caelum Forder, Suman Deb, Ahsen Tahir, Roman J. Georgio, Peter Carroll, Zekun Guo

▸Proposed a workflow-free, information-flow-orchestrated multi-agent system using agent-to-agent (A2A) natural-language communication.
▸Introduced a centralized information-flow orchestrator for dynamic task monitoring and adaptive agent coordination.
▸Achieved 63.64% pass@1 on the GAIA benchmark, outperforming the workflow-based OWL baseline by +8.49% at comparable token usage.
▸Demonstrated improved robustness on edge cases and long-horizon task decomposition.

Multi-Agent,A2A Protocol,GAIA Benchmark,Agent Orchestration

Beyond Rule-Based Workflows: An Information-Flow-Orchestrated Multi-Agent Paradigm via Agent-to-Agent Communication

★

Accepted atNeurIPS '25 · LAW Workshop

2025 · arXiv · AcceptedNeurIPS LAW Workshop

Anemoi: A Semi-Centralized Multi-Agent System Based on Agent-to-Agent Communication

Xinxing Ren, Caelum Forder, Qianbo Zang, Ahsen Tahir, Roman J. Georgio, Suman Deb, Peter Carroll, Önder Gürcan, Zekun Guo

▸Proposed Anemoi, a semi-centralized multi-agent system leveraging Agent-to-Agent (A2A) communication to reduce reliance on a single planner.
▸Enabled structured inter-agent collaboration via an MCP server with real-time monitoring and adaptive plan refinement.
▸Reached 52.73% accuracy on the GAIA benchmark using a small planner LLM (GPT-4.1-mini).
▸Outperformed the OWL baseline by +9.09% under identical LLM settings, with improved scalability and robustness.

Semi-Centralized,A2A Protocol,MCP,GAIA Benchmark,NeurIPS

Anemoi: A Semi-Centralized Multi-Agent System Based on Agent-to-Agent Communication

Experience

Jun 2026 — Present

Research Intern · Canada

AI Research Intern

University of New Brunswick — Fredericton, Canada

▸One of the top applicants chosen through Canada's Mitacs Globalink Research Internship, a fully funded and highly competitive international program attracting undergraduates from over 20 partner countries through a rigorous multi-stage screening and faculty matching process.
▸Conducting applied research benchmarking LLM performance on coding tasks, evaluating model capabilities across correctness, efficiency, and generalization under the supervision of Canadian university faculty.

LLM BenchmarkingCode GenerationPythonEvaluationResearch

Feb 2026 — Present

Co-Founder

Velton — GEO Agency — Karachi · Global clients

▸Founded a GEO-focused agency targeting luxury aviation, yacht charter, and premium automotive brands, building client visibility across AI-powered search platforms.
▸Building automated content and outreach workflows to drive measurable visibility outcomes for clients across AI-powered search platforms.

FounderGEOGTMLuxury brands

Feb 2025 — Present

AI Engineer

CoralOS — Remote · London, UK

▸Shipped agent-to-agent communication infrastructure used in production multi-agent systems, with structured protocols for task orchestration and real-world testing across diverse use cases.
▸Built and evaluated interoperability across multiple LLM agent frameworks, including prompt design, dynamic role delegation, and structured output pipelines.
▸Built autonomous agents enabling protocol-driven communication and reliable automation across distributed environments, as part of production-facing multi-agent infrastructure.

MCP serversA2A