Harshanath's Digital
Atmosphere

Routine monitoring processes were previously paper-based, highly time-consuming, difficult to validate, and heavily dependent on manual monthly reporting cycles. Earlier attempts to digitize using Microsoft Forms were unsuccessful due to unstable network connectivity in rural implementation areas, forcing the organization to return to hard-copy field monitoring methods. This operational inefficiency created delays in reporting, increased administrative burden, reduced field productivity, and severely limited management visibility over live monitoring conditions.

A relational data ecosystem was designed using twenty-two SharePoint Lists corresponding to twenty-two MEAL tools used by the organization. These lists were connected to a Power Apps mobile application capable of offline field data collection using local storage and synchronization logic — officers can now collect monitoring data without network availability and synchronize records automatically once connectivity becomes available.

Data validation mechanisms were implemented directly within the application layer to improve data quality and reduce reporting inconsistencies. Beneficiary geotagging and mapping functionality were integrated into the application ecosystem for location-based monitoring and field-level traceability.

All collected data is connected directly into live Power BI dashboards, eliminating the need for traditional manual monthly reporting workflows entirely. Management can now access live monitoring information through mobile-accessible dashboards, while MEAL officers can redirect more time toward actual field monitoring rather than administrative report preparation. The entire ecosystem was developed with zero additional software cost — the organization already operated within the Microsoft 365 environment.

Although monitoring findings and dashboard insights became available through MEAL HUB, the identified field issues did not automatically reach the implementation-level officers responsible for corrective action. This created a structural disconnect between monitoring observations and operational response mechanisms — findings existed in the system but produced no direct accountability chain at the field level.

A separate SharePoint environment was intentionally created to isolate implementation workflows from core MEAL monitoring databases, preserving data security and access control boundaries. Relational database structures were implemented using SharePoint Lists connected through primary and foreign key logic.

Power Automate workflows were designed to extract monitoring findings from MEAL HUB datasets, transform the records into issue-management structures, and load them automatically into the FIMS operational environment — creating a continuous ETL pipeline between the monitoring and accountability layers.

Officers logging into the system using their Microsoft 365 organizational accounts can view issues specifically related to their assigned activities and operational responsibilities. Officers can review issues, update resolution progress, mark issues as resolved, or escalate unresolved items when necessary. District managers and regional managers have supervisory visibility and can reassign issue responsibility between officers when staffing changes or operational restructuring occurs.

Officers can access implementation guidelines and monitoring expectations directly within the application before conducting activities — enabling preventive quality assurance rather than only post-monitoring correction. This functionality was designed to reduce operational mistakes before they occur and strengthen implementation quality at the field level by embedding institutional knowledge directly into the activity workflow.

MEAL HUB handles routine field monitoring, and FIMS routes the issues it surfaces to the implementation staff responsible for correcting them. The FIMS Dashboard closes the loop: it is built directly on the FIMS database and turns raw issue-resolution activity into insights that both field officers and management can act on immediately, rather than waiting for a monthly report.

The dashboard is built with Streamlit, with its source maintained in a private GitHub repository and deployed as a private application on Streamlit Community Cloud. Access is restricted to the organization — no external party can reach the dashboard or the data behind it. As with MEAL HUB and FIMS, this was achieved at zero additional organizational cost, completing the full digitalization of the MEAL routine-monitoring cycle at Berendina Development Services.

The dashboard has become a genuine motivator for field staff — officers can watch their own issue-resolution progress live, and the system ranks the quickest and most consistent problem-solvers. This has introduced a healthy sense of competition around resolving issues on time, while removing the need for any additional manual reporting to management. Everything management needs is already digital and current.

The version embedded below runs on sample data — beneficiary names, officer names, and GPS coordinates have been altered and do not reflect real organizational records.

Berendina Development Services operates across three regions. This site was built to publish case stories from the NC Region — the region I monitor — where six officers currently implement projects. Officers submit stories from the field, and I review and publish them here, building a growing, public record of impact over the course of the year.

Sri Lanka's development sector relies on weather-sensitive planning, while the traditional bimodal monsoon calendar has become highly unreliable due to recent climate change. Available local and global forecasting services provide only a 3–5 day reliable horizon, without district-level operational thresholds for activity planning. This leaves development agencies and NGOs vulnerable when identifying safe windows for weather-sensitive interventions such as tea cultivation, infrastructure construction, livestock distribution, and community training.

This project bridges that gap by designing, building, validating, and deploying a Weather Decision Support System (W-DSS). The system uses a district-aware multivariate LSTM network with attention as its forecasting core, trained on eleven years of daily weather data (2010–2023) from thirty Sri Lankan cities, mapped during feature engineering into seventeen districts. In a single forward pass, the model forecasts key weather variables thirty days ahead and derives three rainfall categories — Safe, Caution, and Risk — avoiding the compounding error typical of autoregressive single-step forecasting. A rules-based recommendation engine then converts raw weather output into four activity-specific, literature-based recommendations, reporting the model's confidence and the specific evidence threshold it relied on.

On an unseen test period, day-one rainfall category classification accuracy was 66.56%, settling to a stable plateau of 52–55% by day thirty (52.70% at day 30) — consistently well above the approximate climatological baseline of 35%, and in line with published literature on the limits of tropical rainfall predictability. Dry Zone districts outperformed Central Highland districts by a meaningful margin, supporting the district-embedding approach used in the model design.

The completed W-DSS is deployed as a publicly accessible web application on Streamlit Community Cloud. The project demonstrates that a forecasting model paired with an interpretable, evidence-based, and honestly-bounded translation layer can be genuinely useful for decision-making in a climatically difficult tropical setting.

Vesak Lanterns reimagines a familiar ritual — lighting a lamp — as a small, shared digital experience. Visitors light a lamp and pair it with a positive thought, joining a growing, participatory display built by the community rather than by any single person. On the last Vesak Poya Day, many people lit a lamp with a thought of their own, showing real appetite for celebrating Vesak in a new way alongside the traditional one.

Anyone can view the lamps already lit, each carrying its own thought. New lamps can only be lit during the Vesak period itself — the site is intentionally locked outside of that window and will reopen for the next Vesak Poya Day, keeping the ritual tied to its proper occasion rather than turning it into an always-on novelty.