BU⚡️

Tweet about renewable energy, politics, economy, & emerging technologies in Africa. LFC ❤#YNWA

[BU⚡️ @ubahbenson]

@Ssaasquatch I tap from this blessings and proclaim that 2025 will be that year for me!

[BU⚡️ @ubahbenson]

@Morris_Monye Happy birthday Chief!
More blessings.

[BU⚡️ @ubahbenson]

@azeezibrahim3 @TechDJKella Amen 🙏 it is well

[BU⚡️ @ubahbenson]

@christinuella @Ksmd2026 Congratulations doc!

[BU⚡️ @ubahbenson]

God, I need $100k.

[Osaretin Victor Asemota @asemota]

God, I need $400k.

[Osaretin Victor Asemota @asemota]

God, I need $400k.

[BU⚡️ @ubahbenson]

@OluwaBishopBaba Happy birthday bishop 🎉🎉 more blessings

[BU⚡️ @ubahbenson]

@johnternus Goodluck!!

[BU⚡️ @ubahbenson]

@cindy_blog Amen 🙏

[BU⚡️ @ubahbenson]

@Kolettegene Congratulations @Kolettegene

[Mundus @mundus01]

One of the biggest challenges in building https://t.co/T1C9aFSkry, beyond setting up the data pipeline, was OCR.

Polling-unit result sheets are handwritten, then photographed and uploaded to IReV. The handwriting, image quality, lighting, angles, and compression meant that traditional OCR tools simply weren’t reliable enough.

We tested several of the leading LLMs and eventually settled on a dual-model consensus approach: two different models independently read each result sheet, and a vote total is accepted only when both models agree.

Even with that safeguard, we knew there would still be a small chance of incorrectly reading some polling-unit results.

Before Osun, we tested the system using results from Ekiti and Anambra, and the results gave us enough confidence to deploy it at scale.

Now that INEC has announced the official results for all 30 LGAs in Osun, we can finally compare the official statewide collation with what Mundx independently produced from polling-unit results uploaded to IReV.

The result is encouraging:

Mundx’s polling-unit tally was about 95.68% accurate compared with the final INEC statewide results, despite having only 97.77% of polling units included in its count.


That means an automated, LLM-based system was able to get remarkably close to the final official result while working from an incomplete set of polling-unit uploads and doing so much faster and more cost-efficiently than manual collation.

This has reinforced something we strongly believed while building the system: LLM-based OCR is incredibly powerful, especially for documents that traditional OCR struggles with.

There is still more work to do. We will continue strengthening the validation layers, improving how the system handles difficult result sheets and edge cases, and making the platform even more resilient ahead of 2027.

Thank you to everyone who followed the project, tested the platform, shared feedback, and sent kind words throughout the election.

Congratulations Governor Adeleke.

[BU⚡️ @ubahbenson]

@mundus01 Great job Mundus!

[BU⚡️ @ubahbenson]

@_belikebaddy Congratulations on your successful surgery! Wishing you a quick recovery.

[BU⚡️ @ubahbenson]

@crazy_kennar 😂😂😂 I came to rep 9ja. BAT juju too strong!

[BU⚡️ @ubahbenson]

@aproko_doctor Call Ambulance like say e easy to get one for critical moments.

[Olumide Adesina @olumidecapital]

Many people you assume to be rich on social media are living on rent life style, ill-gotten wealth or debts. Focus on your lane and do the work

[BU⚡️ @ubahbenson]

@Favour_Q @Onomz_D1 lol 😂 valid question

[BU⚡️ @ubahbenson]

@NigeriaStories @sinzubaba They have practically raised the Standard like their name 😊

[BU⚡️ @ubahbenson]

@Adikastakes Happy birthday Adika. #YNWA

[BU⚡️ @ubahbenson]

@tosinolaseinde True!