-
-
The UI
-
the settings
Inspiration
The inspiration struck when I observed people in my circle relying on AI for critical, life-impacting advice (such as medical inquiries). I realized the inherent danger of AI "hallucination" and its dangerous tendency towards "sycophancy" (compromising facts to please the user). I decided to build Mizan to ruthlessly suppress these flaws. The core philosophy was inspired by the meticulous process of coffee dripping: distilling raw, chaotic information through complex, proprietary layers to extract only the purest, most accurate essence.
What it does
MEEZAN acts as a rigorous truth-filter. It delivers hyper-accurate responses completely devoid of hallucinations, biases, or user-pleasing tendencies. It features an innovative "Tri-Output" system, providing three distinct adaptive responses—either customized by the user's domain or intelligently auto-generated by the framework. Beyond its logical backend, MEEZAN features a highly refined, visually balanced, and eye-friendly UI, avoiding harsh or dull colors to ensure optimal user experience. The architecture is versatile and adaptable to virtually any professional field that does not strictly require pure human intervention.
How I built it
For the frontend, I designed a web-based, eye-friendly interface using HTML and standard web technologies to ensure a smooth and accessible user experience. On the backend, I built a modular, multi-agent architecture utilizing a proprietary logic-gating system. The core focus was on establishing strict communication protocols between the generative and auditing paths to ensure unbiased information distillation.
Challenges I ran into
The most complex technical hurdle was engineering the dynamic "auto-generate" system, ensuring the agents seamlessly and accurately adapt whenever the user shifts the operational domain. Additionally, perfecting the UI to achieve visual equilibrium without compromising the user experience required meticulous iteration. Finally, as a solo developer, I carried the entire architectural, design, and logical workload alone, which meant solving every bottleneck and making every critical decision without a team to brainstorm or divide the labor with.
Accomplishments that I proud of
I am incredibly proud of benchmarking MEEZAN against an industry-leading, massive AI and seeing it triumph. Where the famous model exhibited sycophancy and altered its tone to appease the prompt, MEEZAN remained coldly objective and ignored user-pleasing entirely. It responds with the authoritative precision of a seasoned veteran in any given field, acting as a purpose-built industrial tool rather than a soulless chatbot. I am equally proud of achieving this clinical accuracy while maintaining an astonishingly fast response time, masking the massive computational logic happening behind the scenes.
What I learned
I learned advanced architectural strategies to neutralize AI hallucination and bias—a fundamental flaw that even the world's biggest tech giants are currently battling. Realizing that I, as a solo developer, could engineer a functional framework to solve this core industry problem was a profound learning experience and a massive milestone.
What's next for MEEZAN
My ultimate vision is to see MEEZAN adopted by enterprise sectors, fully scaling the architecture to its "Gold Standard" of 1000+ agents to unleash its true processing power. In the near term, I aim to secure cloud compute sponsorships to host these agents efficiently and to integrate specialized, zero-hallucination base models for the framework's core auditing nodes.
Bonus Blog Post: Securing Sovereign AI with Auth0 Token Vault
Building Meezan: A Journey into Zero-Trust AI Identity
Integrating high-stakes AI with external APIs has always been a "security nightmare" for developers. When I started building Meezan, my biggest challenge was: How can a sovereign, local-first AI framework interact with the outside world without ever touching or leaking sensitive OAuth tokens?
This is where the Auth0 Token Vault became the game-changer for my project. In traditional AI architectures, you often have to pass raw API keys directly to the LLM or handle complex token refresh logic within the application code—exposing the system to prompt injection and credential theft. By implementing Auth0 Token Vault, I was able to completely abstract the identity layer.
The Vault acted as a secure broker. My AI agents now simply request an action, and the Vault handles the token injection at the edge. The "Meezan" agents never see the raw credentials. This allowed me to maintain a 100% Zero-Trust architecture. For a sovereign framework that prioritizes data privacy above all else, this integration wasn't just a feature; it was the missing piece of the puzzle.
Auth0’s ecosystem allowed me to scale Meezan from a local reasoning engine to a secure, enterprise-ready tool that respects user consent and identity. Seeing the "Majority Consensus" agent approve a transaction based on a secure identity check was the most rewarding part of this hackathon. This journey proved that with the right tools like Token Vault, we can finally build AI that is both incredibly powerful and fundamentally secure.
Log in or sign up for Devpost to join the conversation.