Hungerless

Inspiration

Theme: Adaptation & Resilience / Climate Technology & Infrastructure Living and building tech in Nairobi, the devastating cycle of reactive humanitarian aid is impossible to ignore. Every year, predictable climate patterns trigger droughts, yet interventions usually only begin after a famine is declared. By then, logistics costs are astronomical, and the human toll is already severe. I wanted to shift this paradigm from reactive emergency response to anticipatory action. The goal was to build a system that uses historical data and predictive modeling to identify deficits and pre-calculate the exact logistical response required months in advance.

What it does

Hungerless is an Anticipatory Action & Logistics Engine. It processes historical climate, livestock, and seasonal forecast data to predict county-level food deficits or surpluses. Once a risk is identified, the platform dynamically maps out the most cost-effective supply chain routes to move surplus food to deficit zones. Finally, it acts as an AI strategist, utilizing DeepSeek-V3.2 to generate executive summaries, ROI analyses, and security assessments based strictly on the calculated ground truth. By pre-positioning resources before a crisis peaks, Hungerless centers Equity & Justice, ensuring that historically marginalized communities in Arid and Semi-Arid Lands (ASAL) aren't left waiting for slow, reactive aid to trickle down.

How we built it

Drawing on experience building robust data analysis tools and dashboards, I engineered Hungerless with a strict "Math-First" architecture to prevent AI hallucinations.

1. The Prediction Engine: We built a deterministic Python backend using Pandas and NumPy to process CSV datasets. Yield drops are calculated using a non-linear sensitivity model: $$\text{Yield}{pred} = \text{Yield}{base} \times \left(1 + 1.2 \times \frac{\text{Rain}{forecast} - \text{Rain}{avg}}{\text{Rain}_{avg}}\right)$$

2. The Logistics Engine: We calculate exact route distances between counties using the Haversine formula on raw GPS coordinates: $$d = 2R \arcsin\left(\sqrt{\sin^2\left(\frac{\Delta\varphi}{2}\right) + \cos\varphi_1 \cos\varphi_2 \sin^2\left(\frac{\Delta\lambda}{2}\right)}\right)$$ This data is then run against real-world OSINT metrics—specifically March 2026 EPRA diesel prices and KeNHA axle load limits—to generate precise budgets for different commercial fleets (e.g., Isuzu NPR, Scania 6-Axle).

3. The Reasoning Layer: We integrated the Featherless AI serverless platform to power the DeepSeek-V3.2 model. Instead of asking the LLM to guess logistics, we feed it the deterministic JSON output from our math engines. The AI is used purely for high-level reasoning and drafting stakeholder reports.

The entire system is wrapped in a Streamlit frontend with Folium for interactive geospatial mapping.

Challenges we ran into

The biggest challenge was bridging the gap between generative AI and deterministic logistics. In humanitarian aid, an LLM hallucinating a route distance or a fuel budget can cause catastrophic operational failures. We had to strictly isolate the AI, ensuring it acted only as an interpreter of the hard data, rather than a generator of facts. Additionally, sourcing and integrating accurate local metrics—like standard Kenyan trucking capacities and live fuel rates—was complex but necessary for the platform to be viable in the real world.

Accomplishments that we're proud of

I am extremely proud of the Hybrid Intelligence Architecture. By successfully separating the "Calculation Layer" from the "Reasoning Layer," Hungerless proves that LLMs can be safely deployed in high-stakes environments if they are tethered to solid mathematics. Deploying DeepSeek-V3.2 via Featherless API was seamless and provided incredible strategic depth to the final product.

What we learned

Building this reinforced that the future of applied AI is not just better prompts, but better guardrails. Grounding an LLM in mathematical models completely changes its utility, turning it from a chatbot into a reliable operational strategist. I also gained deep insights into the actual mechanics of East African supply chains and the massive financial leverage that anticipatory action provides over reactive airlifts.

What's next for Hungerless

The immediate next step is expanding the geospatial mapping capabilities to track real-time hazards like floods and impassable roads that disrupt supply chains. We also plan to integrate a Two-Eyed Seeing approach: complementing automated API feeds from meteorological organizations (like NOAA and the Kenya Meteorological Department) with crowdsourced, indigenous ecological knowledge from local pastoralists. This ensures our deterministic models are continually ground-truthed by the communities living on the front lines of climate change.