ZannixLabs pioneers a new AI Agentic Research Paradigm—transforming continuous real-world telemetry lakehouses into peer-ready preprints, mathematical proofs, and open-science observatories at machine speed.
// 01 — THE PIONEERING PROCESS
By shifting from manual data wrangling to automated agentic intelligence, ZannixLabs formalizes empirical hypotheses, computes statistical proofs, and compiles camera-ready preprints in days rather than quarters.
Automated high-frequency telemetry pipelines ingest millions of real-world rows—API gateway tokens, ADS-B transponder state vectors, sports ball trajectories, and crowdsourced battle logs—into an immutable, high-speed analytical lakehouse.
In-process zero-copy vectorized DuckDB execution
Automated schema verification and 29-point quality check suites
Hourly scheduled synchronization without manual data wrangling
// 02 — FLAGSHIP RESEARCH
Three camera-ready studies spanning AI Gateway Economics, Leaderboard Biases, and Aerospace Airspace Telemetry.
A study of 646 Trillion words: free open models handle 66% of queries, but closed giants take 60% of revenues
We analyzed 646 Trillion words generated across over 1,000 AI models worldwide to answer a simple question: Where does the money in AI actually go? We discovered two big surprises: First, free open-source models handle nearly two-thirds of worldwide usage, but closed corporate giants pocket 60% of all revenues. Second, when developers use AI coding assistants on real projects (50+ questions), costs don't increase steadily—they explode by 32 times because the software re-reads the entire chat history on every single turn.
Empirically Verified Findings:
Testing 3.7 million human votes to uncover why chatbots win votes even when their answers are wrong
Popular AI ranking leaderboards ask everyday users to compare two anonymous chatbot answers and pick the winner. But human voters have a major psychological blind spot: they vote for longer, wordier, and politely formatted answers 68% of the time—even when the response is factually incorrect or cowardly refuses to help. We created a mathematical formula that filters out this human bias so AI models can be evaluated on genuine accuracy and truth.
Empirically Verified Findings:
Tracking 4.1 million real flight signals across Europe to uncover predictable airport runway traffic jams
When your flight is put in a 'holding pattern' circling in the sky before landing, it's not bad luck—it's a predictable traffic jam. By analyzing 4.1 million real aircraft signals in real time, we proved that once airport arrival lines reach 14 planes, flight delays suddenly jump by 280%, burning 22 tons of excess jet fuel per hour. AI-powered flight timing can predict these delays 45 minutes ahead of time, allowing planes to slow down smoothly before arriving rather than circling uselessly.
Empirically Verified Findings:
// 03 — LAB OBSERVATORY
Science must be verifiable. Explore interactive visualizations generated directly from our lakehouse telemetry pipelines.
Real-world data ingested and synthesized directly through Zannix analytical pipelines.
Analyzed over 646.2T tokens across 1,091 model endpoints.
Figure 1: Telemetry snapshot • Cyan data points indicate non-dominated Pareto frontier models.
ZannixLabs was founded by Ajmal Baba with a clear mandate: replace slow, siloed, sponsor-driven research with high-throughput, agent-driven scientific synthesis.
Ajmal Baba
Founder & Chief Scientist • Independent Quantitative Researcher
Leading autonomous agentic empirical research across AI, aerospace, and quantitative dynamics.
Deterministic DuckDB schemas, executable SQL notebooks, and public telemetry.
Complete institutional independence. No paid vendor endorsements.
Formal LaTeX math, Welch two-sample t-tests, and Pareto frontiers.
High-frequency data streaming directly into analytical lakehouse stores.