Build For Bharat Fellowship @ Rural Development and Panchayat Raj Department, Government of Karnataka
Can local data and simple tools help the government work?
Yes. When organized, observed, and acted upon.
I spent the summer of 2025 exploring, organizing, and building upon rural data in Karnataka, India. Building tech solutions for the public sector, directly with the government, was something I could hardly have anticipated.
The Rural Development and Panchayat Raj Department, Government of Karnataka, has devised its IT system, known as Panchatantra. This platform has various modules and areas for recording and conveying information. These are Meeting Management, Revenue Collection, Citizen Services, Planning, Panchamitra, etc..
I, along with my fellow teammates Veda Arvind and Shanu Priya Garg, had to work on 3 projects for the summer. Out of the three, my work was related to:
- Meeting Management and Minutes Analysis: The RDPR Department has standardized the frequency of various types of meetings held in Grama Panchayats. Panchayat officials are responsible for digitally recording the meeting minutes, including agendas, participants, discussions, decisions, attendance, and other relevant details. Our task was to work on the Analysis of these meeting minutes and research about the meeting management process to identify the pain points and areas of improvement. This included programmatically organising the data, and also visiting the Grama Panchayats in person to better customize the interface, results, and process of recording.
- AI Radio/Video: With the Panchatantra platform and other government sites, the citizens could have access to all the government works, plans, schemes, progress, budgets, etc. in the Panchayat. They can read the data on their phone, but basic, digital, and data literacy are big factors to consider in this process. The people seem to be more comfortable listening and viewing the updates, instead of reading. For this, the Government has decided to release AI-generated, realistic audio and video episodes for the citizens in their local language. My work was to review the existing manual pipeline and work towards its automation.
I worked majorly on the Meeting Minutes Analysis for the summer, and looking forward to (and working on!) ways to contribute further in both projects. For these projects, I had to work on Data pipelines, HUGE datasets, and convey the organized results to higher officials in a precise and concise manner.
For the projects , initially, I picked up Python and its tools like Pandas, Matplotlib, GeoPandas, Jupyter Notebook, etc.. I thought these were simple at first, but beneath the simplicity, there was much to know. They work really good together, and the best thing is that all are open-sourced products. Beyond these Python tools, I also had the opportunity to work with various Large Language Models (LLMs) and their associated tools. I experimented with multiple popular LLMs, as well as their specialized products such as text-to-speech systems, script-generation models, and speech-to-text solutions. This work involved building practical workflows and integrating APIs into real data pipelines. While working with large datasets, I implemented techniques like rate limiting, batch processing, request optimization, caching, and error handling to make the process more reliable and efficient. I also learnt and spent time on cost optimization strategies, ensuring reliability in large-scale work. This experience gave me a deeper understanding of how to think beyond simply calling an API, towards building scalable, production-ready AI solutions.
To kickstart things, we had an introductory meeting with the then-Director, eGovernance, RDPR. In the meeting, Ma'am told us to read the 73rd Constitutional Amendment Act (1992) and introduced us to the workings of the department and its mission. Then, I was given access to the initial data constituting various information about the meetings happening in all the Grama Panchayats of Karnataka state, for the last 3 months. This data was, well, huge and difficult to manage. There were multiple files, some containing one information, and some containing the other. These files had some connecting information to relate to each other.
Initially, I was struggling to even get the file encoding right, to parse both the Kannada and English texts. Finally, UTF-16-BE it was! Slowly and steadily, I was able to extract some exploratory data from the files. Then, more results started to come out, and I, along with mentors at [Bharat Digital](https://www.bharatdigital.io/), started to plan our pilot presentation with the department, in order to get more data to explore. The yearly *Ideal vs Actual frequencies* of all the *types of meetings* happening in the Grama Panchayats, the *Agendas discussed* on average per meeting, the *length* of these discussions, and some more Panchayat-specific and Meeting-specific data were derived for all the Panchayats and Districts. Apart from Python tools, LLMs were also utilised in the process, parsing thousands of queries, hour-long code runs, and lots of rate-limiting on API calls.
We plotted all the results as graphs and used color coding to indicate *favourable vs unfavourable* results. Then came the presentation day, and we were able to spark conversations around the present state of many processes through our numbers. The districts following all the regulations, the ones lagging, Panchayats that need special attention, etc.. The then-Director ma'am was impressed with the results and guided us for further study. She suggested the use of maps instead of graphs to better show geographical data, more parameters of quality we could explore, and promised access to more data.
I received the data for the previous year, and this time it was even larger: millions of entries spread across multiple CSVs - so large that MS Excel could not open them fully, and parsing them with code took nearly ten hours. Along with the Grama Panchayat records, I also received Panchayat Shapefiles, which were essential in the final step of visualizing our results as heatmaps across Karnataka. With this richer dataset, I was able to derive results across a wider range of parameters. Beyond tracking meeting frequencies and compliance with regulations, we (I, with my mentors at Bharat Digital) identified the specific topics being discussed and evaluated the quality of these discussions.
After reproducing all the results from the earlier presentation (and even more!) for the full year, we shifted our focus to analyze the content itself. Together, we brainstormed metrics that would be both meaningful and practical for the final presentation. One of these was *categorizing meeting agendas*. We started by sampling agendas at random, extracting unique categories, and then grouping them into broader types. Each agenda was then mapped to one of these types, helping us gauge what was being discussed and how often, quantitatively. The next parameter was *Specificity*: whether an agenda item was discussed in concrete terms or just as a generic issue. This allowed us to assess not only what was discussed but also how well it was discussed. Finally, we extracted recurring *themes* and mapped them visually, which opened the door to a more standardized approach to recording minutes of meetings. This process could eventually support Panchayat officials by providing default agenda templates, making discussions more structured and actionable.
The analysis culminated in *heatmaps* built with the *Grama Panchayat Shapefiles*, ready for presentation. We had the honour of presenting our results in front of [Uma Mahadevan Dasgupta](https://www.linkedin.com/in/uma-mahadevan-dasgupta-287223179/) ma'am (Additional Chief Secretary and Development Commissioner, Government of Karnataka), Arundhati Chandrashekhar ma'am (Director, RDWSD, Karnataka), and Rahul Ratnam Pandey sir (Director, e-Governance, RDPR-GoK). The officials were very impressed by the work, and the amount of time they gave us was really invaluable. We got to ask our questions about governance from the very people who are running it. Some of the *busiest people in the state* were interacting with young fellows about the future of our people. The experience can't be described in words.
"If it gets 1 hour in extracting data, then we have to look at that extracted data for much more than that, to get some actionable insights. We convey Insights, not just Data". This is what I have learned through this experience. At first, I was just thinking of it the engineer's way - we present them with numbers, with data, and make them realise how much it is all worth it. It's after our pilot presentation and discussion with my mentors Harsh Nisar, Richa Marwah, Sabhyata Jain, and Rukmini S that I got to know, we present them with their language, using our data and numbers.
This was all about the project. This was all about me transitioning from *Development to Data Engineering, Product thinking, and Applied AI Engineering*. Thank you for reading. I hope this made our methods and findings clear - and showed how small, data-driven changes can improve governance on the ground. I remain grateful to my fellow teammates (Veda Arvind and Shanu Priya Garg), the entire Bharat Digital team, and the Rural Development & Panchayat Raj Department team for their guidance and help whenever needed.
Onwards and Upwards🚀