DEV Community: Zoheb Malik

We built a new AI Topology to bypass the Transformer bottleneck. Here are our first benchmark results.

Zoheb Malik — Mon, 22 Jun 2026 16:51:55 +0000

If you’ve been following the AI space, you know we are hitting a physical compute ceiling. Standard autoregressive LLMs (like GPT or Claude) are incredible, but under the hood, they are essentially performing highly-educated linear guessing. They require massive, power-hungry data centers just to calculate the next token.

At Trijna Labs, our engineering team decided to stop trying to optimize transformers and instead try to build a completely new neural architecture from the ground up.

We wanted to see if we could build continuous-learning neural topologies that utilize topological entropy routing—essentially allowing the model to dynamically calculate the exact complexity of a query and only spin up the necessary weights, drastically reducing GPU overhead while preserving logic.

We call our primary topologies the ARS Engine (Algorithmic Resonance Sequence) and the OSM Engine (Operational Structural Matrix).

After months of mathematical dead-ends and late-night debugging, we finally got the engines stable enough to run through the official EleutherAI lm_eval harness. We decided to test them on GSM8K (for raw math) and the LiveBench framework (for abstract reasoning).

Honestly, we were nervous to see how a custom architecture would hold up against the massive parameter counts of standard LLMs. But the numbers came back, and they kind of blew our minds.

📊 The Benchmark Results

LiveBench (Overall Intelligence & Reasoning) We pushed our ARS Engine through LiveBench, and it achieved an 87.5 overall average. The most shocking part was its abstract reasoning score, which hit an incredible 93.9. For context on pure logic and spatial tasks, this actually pushes it past the baseline of GPT-4o and Claude 3.5 Sonnet. Because the ARS topology uses geometric spatial routing rather than linear guessing, it practically eliminates standard spatial hallucinations.
GSM8K (Math Word Problems) We ran our OSM Engine (which is tuned specifically for matrix stabilization) through the GSM8K math benchmark using 5-Shot Exact Match. It peaked at 85.06%, proving that non-transformer continuous-learning models can handle complex, multi-step math without memory degradation.

🛠️ How We Did It
Building this wasn't easy. A huge challenge was preventing catastrophic forgetting during continuous training (since we aren't just doing massive pre-training runs). We solved this using a Riemannian Metric Constraint to "freeze" vital parameters based on their importance, geometrically preserving established memory pathways.

🤝 We'd Love Your Feedback
We know we still have a massive mountain to climb to scale these topologies globally, but seeing a non-transformer architecture hit these numbers on local, highly-constrained hardware feels like a huge validation of our physics-based approach.

If you are an AI researcher, mathematician, or just a massive nerd for neural architecture, we uploaded our full methodology, exact dataset hashes, and reproducibility commands to our dev log.

You can read the full breakdown here: Trijna Labs Dev Log

We are a small team trying to do something insanely difficult. If you have any architectural advice, critiques on our math, or brutal feedback, we’d honestly love to hear it in the comments.

Let's discuss! What do you guys think the post-transformer era of AI will look like?

Uracil Labs successfully integrated Uracil Chain in Uracil AI

Zoheb Malik — Fri, 08 May 2026 17:38:56 +0000

Uracil Labs successfully completed integration of Blockchain which runs on AI.

Building a Quantum-Resistant Blockchain from Scratch in Rust: What I Learned

Zoheb Malik — Thu, 02 Apr 2026 18:53:43 +0000

I built a complete quantum-resistant blockchain from scratch in Rust. No frameworks. No shortcuts. 50,000+ lines of code. 106 adversarial tests passed. A 13-step ZK audit completed.

Here's why I did it and what I learned.

The Problem
Most blockchains today have fundamental flaws that we've just accepted:

Assets have no identity
Your wallet says "100 tokens" — but which tokens? Where did they come from? Who owned them before you? There's no way to know. Balances are just numbers.
Private keys are a single point of failure
One stolen key = everything gone. No built-in 2FA. No key evolution. No recovery mechanism.
Consensus forces trade-offs
PoW is secure but slow. PoS is fast but capital-concentrated. You have to choose.
Mining is wasteful
We burn massive energy for pure competition. No productive output. No scientific contribution.

What I Built — Uracil Chain
Ghost Chain — Serial-Based Asset Tracking

Every token is a unique serial number, organized into efficient ranges.

Instead of storing 1 million NFTs as 1 million separate accounts (~200 MB), Ghost Chain compresses them into ~100 KB — a 99.95% storage reduction.

Every serial has a full history: creation, every transfer, current owner. Full forensic auditability.

Guardian Keys v2 — Dual-Factor Auto-Rotating Security

True 2FA for blockchain wallets:

Independent seed generated once, displayed once, stored only locally

Never touches the blockchain

Both private key and seed required to send transactions

Keys auto-rotate every transaction

A stolen private key alone is useless. You need the seed too.

PoAuth — Weighted Multi-Sig with Falcon Signatures

Quantum-resistant governance for DAOs and enterprises:

Falcon signatures (NIST PQC finalist) — 36–44ms signing, ~5ms verification

Weighted thresholds: CEO weight 5, CFO weight 3, CTO weight 2

Multi-sig quorum with quantum security

PoUW — Proof-of-Useful-Work

The first blockchain that rewards scientific computation.

Nova-based ZK proofs for protein folding

Validates against Ramachandran constraints (real biophysics)

3 residues in ~4.0s, scales linearly

13-step audit passed

Mining becomes productive. Your compute power funds science, not just hashing.

Dual-Chain Architecture

Separates execution from finality:

Live Chain: instant validation (<1ms), reversible

Archive Chain: secure consensus, final, irreversible

Users get instant feedback. Security stays strong. No trade-off.

Performance Benchmarks
Metric Result
Live chain throughput 16,753 TPS
Ghost balance query 2.28 µs
Falcon signing 36–44 ms
Protein folding (3 residues) ~4.0s
Testing
106 adversarial tests across 11 layers

Deterministic replay, economic attacks, network attacks, consensus attacks, ZK proof validation

All passed

What I Learned

Start with the problem, not the solution
I asked: "What would a blockchain look like if it were designed for asset provenance from day one?" That question drove every design decision.
Range-based compression is powerful
The insight that assets can be stored as ranges rather than individuals seems obvious in retrospect — but nobody was doing it. Sometimes the best solutions are simple ideas executed well.
ZK proofs are ready for production
Nova SNARKs are practical. 4 seconds for 3 residues is usable. With pattern compression (I found 95% repetition across residues), theoretical optimization hits 94%.
Open source changes everything
The project is MIT-licensed. People are cloning it (52 unique cloners in 14 days). The feedback loop is invaluable.

What's Next
Third-party security audit

Deploy components as developer tooling for Ethereum, Solana, and peaq

Build custom ZK compressor (94% theoretical compression)

Launch mainnet

The Code
Everything is here: https://github.com/jcinfosolution-hash/Uracil-Labs

50,000+ lines Rust

Full documentation in /Docs

MIT license