The conference started. And I am almost on time. After the every explicit support for grievance and misconduct reports, the dev rooms and the wide range of the topics

23% women and 1% non binary in the registered participants. There are also lighting talks with open slots open for on spot submissions.

First talk starts already. One of the first GitHub open source grant. On review structure and governance of open source health tech.

From pull requests to Patients by Bodhish Thomas. About building open source beyond ideology and for accountibitily.

No downtime, and deployments can't be taken down. PR to release to deployment to training to workflow to final user. The presentation has slop, but luckily the Bodhish's delivery does not.

Started as COVID war room, government adopted it in Kerala. From 2023, Open Healthcare Network and 10 bed ICU and built ICU management software. Including TeleICU.

Kerala's palliative care grid for home based health care provision and palliative care by nurses. Multiple states now using it.

Large enough scale with multi stage environment mainly around and because of the open source tool building. Mainly because infra is owned by gov.

One core layer that can be shared across, Care Core is an os for admin to pharmacy to veternary applications sharing the os.

Idea was to build the primitive, not just features with collection of plugins. A program then it's just a combination of the plugins.

Good contributor onboarding, by letting them learn, get mentored and grow.

Review discipline is very important and high stake. Security expectations are high, with zero access to government owned deployment and data.

Governance is also important and yet community is the core. Without thr builders, it's hard to maintain.

Stewardship underlined that open source are easy to start and deploy but trust factor is the real long term work.

CSR funding no access keeps you stuck for first three years. There is grant wall. FCRA too have similar issues.

There needs to be maintainer remuneration and funding is hard as well. Nobody funds the software that runs hospital and building the institution early is important.

Healthcare does not have the allready existing rails to run like banks or other institutions. Healthcare needs a lot of attention.

Budhish ends the talk by inviting people in to build.

There is going to be a lot of running around during this conference. A lot happening on multiple track. Now in the documentation room for 10 min.

What Documentation Needs to Thrive in the Age of AI Search, waiting for this piece in Room 1. May have to drop out before this starts.

I shall run back to Hall 1 audi for from Institutions to Community. And run back here from another GovTech session on its history in India.

A lot of this is being recorded anyway and most likely be soon available as well. So, I can keep that in mind and be chill. But it help to be on toes for better retrieval too

Looks like is already late. Abandoning this non started session and running down to catch from institutions to communities.

Mohit Tahiliani from COSH, NITK Surathkal talking about FOSS and building open source community. Mainly working in the are of networking. And building COSH.

When Mohit when he started as a faculty, were curious about the effectiveness of the projects and assignment. And students prefer personal paced projects over scheduled assignments.

Giving quality feedback over the manual evaluation of the assignment, rooted in full empathy from students perspective and open mind seems like root intention.

Students club learning were cutting edge and very rigorous on the other hand. Great problems and amazing solutions, not beyond hobby projects.

Given relevance, Prof Mohit then proposed club projects to credit but dropped because it wasn't inclusive for students outside the clubs.

GSoC mentorship opportunities on ns-2 helped Prof Mohit to take inspiration. As a part of the course assignment, the PRs being merged brought more satisfaction and networking.

The invitation to write documentation to take away the fear of coding from students was applauded by the audience.

People started to apply because it was a course assignment too, otherwise GSoC gets sidelined as not my cup of tea.

I shall run back to catch the Bruno's documentation talk, since this is mostly about getting students involved in open source communities through course assignment also funded. Motivation peaks in students.

Back in the documentation dev room, the last skipped lecture is almost ending. One slipping take away from this session is that scope your blog post to one thing.

Padmashree now from Swift who is also GSoC mentee. Plain comments without any IDE features, aren't good enough so syntax highlighter. So the speaker solved that problem.

Reference to DocC by Apple. But these plain comments getting turned to these links has a specific parsing flow.

IndexStore via USR is used to jump to such locations normally. Dual parsing approach, such as Swift and then Markdown parsers are used.

Currently under review and the details are skipped by Padmashree because of the limited time slot.

1045 now. Next one. The Bruno one I am looking forward to along with the emcee here in the hall.

A lot of exodus from the room now. Documentation that drives the adoption and the lessons from that. The title feels like a dev rel version of classic documentation.

Ganesh from Bruno, tells us to prefer humans over AI as our consumers, since writing for human follows the writing for AI as well. Ganesh is dev rel anyway at Bruno.

Ganesh creates content around the product at Bruno. He has startegies, tools and platforms and general playbook to improve the docs.

Ganesh makes a case for starter guide. This is your user entry point, complexity here harms. Show first useful action and use 5-min rule.

Oh. This room is very late. I am here for the GovTech history session. And the Bruno session is still going on 7 min in into that session schedule.

I may have to skip the history of the GovTech for the Techies in Social Sector at 11 in room 3

Ganesh makes a case for a good documentation search. Title and structure something that I can continue to own while using the AI.

The history of GovTech is 10 min session. I can take risk to go late to the Techies in Social Sector session.

Your documentation could be the central venue telling about all your other doc avenues like youtube, community or external sources.

Ganesh is talk is getting stretched well beyond ten min while he arrives at the most important slide about the feedback loop on priorities.

Analytics is the next chapter thay Ganesh opens. I think this is going to get stretched beyond and slow exodus out of the hall has started.

Decided to skip the GovTech now for Techies in Social Sector.

The need for tech volunteers in is social sector by Knadh is being referred in this session. I just came in when it started. Digital public infrastructure Is being talked about. Looks like this a roundtable though.

One of the participant has a challenge about sustainability and focused efforts in once-month meeting.

I might regret being stuck here in this round table. One recent graduate sharing there experience about a very specific use case of health care application and the direct qualitative impact as motivation and satisfaction.

Someone from GIZ Germany here to gauge the open source communities for possible collaboration. Someone from Khoj to create a story telling application for children with an genda to build the skills.

There is also an air of tech solutionalism may not be the solution for all. Someone from Tech4good shares the Agentic ai bridges the gaps in communication in requirement and deployments and building.

This rooms feels like non techies talking about tech and techies talking social impact. It doesn't truly fit or add quality.

One take away for me is to meet people from Tech4Good on there model. There are atleast two of them in the room. Also, follow the GIZ person who stepped out of the room on the scale of the projects they're looking to support

The problem with the social sector is also that there is a lot of baggage and self identity that colludes the political awareness. That also clashes with technology identities.

Now I am at it takes a village to raise a repo. And need of community and meet ups.

The speaker works at the Processing foundation. And talks about the role of the governance and board. And board as inhouse family.

The speaker is also on the board and treasurer, the budget feels very big but not matching to the number of volunteers.

The talk feels more about how it takes the a village but not about how to start a community now in hindsight of the talk.

Next for me is the porting of a voice tech to Rust at 1230. This talk ended.

There is soapbox of kind event now. I can take that meanwhile before the next session.

I could actually watch the live now section on the schedule for better choices for the events to attend.

The Foss United board uses election for legitimacy with representation at the government tables. There are some 10 people contesting this election and out of 25k ballets 2k-3k vote.

I shall rush to the Porting to Rust talk.

Hall 2 just finished the talk on mapping India, this must have been the GIS data viz and questions going on.

It is a packed hall. And slow exodus has started before the next talk. Next one is about voice based AI, Rustvani, a first time speaker Allen George.

Allen George is from SHIS (scalable human interaction systems). Rustvani is under 100mb. Allen starts his talk with global memory shortage.

Allen refers to the proce economics of the chips and the 5x spike is a permanent change. Cheaper software is increasing demand for chips and prices are going up. Even cloud prices will double.

Particularly, India is GPU starved. India, which has 18 languages, needs voice AI. It has a big market 50k crores. Eastern India already has access. Pipecat framework and others are not memory optimized.

Allen thinks in indian context, noice cancellation should be default. Pipecat costs 10rs/min and our market needs 3rs/min.

But allen's scope is beyond india, because india is the most challenging env anyways. If it works here, it works everywhere.

Rustvani is pure rust under 100mb. And it will be alway India first, with integration with Sarvam etc.

Rustvani also integrated Sarvam and OpenAI. Allen shows a recorded demo of the Rustvani over Pipecat with almost 5x performance over lighter machines.

In Voice AI the startup time is important, saving a lot of money on cold instances. Another Hushvani tool, a quantized indian open weight model, is also another product.

Hushvani is a noice cancellation tool. I also already see Gnani bring referred to alongside the Sarvam AI.

Allen invited to fork and contribute in the end.

And there aren't any questions. Except one. Around the pipecat ecosystem compability and Allen says that it is.

Taking a lunch break now. Next is a session at 2pm. I shall hop around the booths to meet folks. After having lunch.

Food was veg and indian. Carbs heavy, oil. React Nexus food was a lot more diverse. I am India Foss has cut prices here.

Data for India booth has interactive survey on Computer Ownership, Computer Use, Domestic Expenses in a demographic and electricity penetration as a primary energy sources in India. It feels interesting booth presence Nd activity.

Paper cranes labs booth had cool P5js and there claim is that it shall not be tough to use P5js in react tooling. I shall explore that.

Juspay seems to have a very popular open source tool hyperswotch for payment integrations and also Prism for individual cases. Under Apache license.

Interacted with the GIZ personel on if there is potential to recieve support from GIZ. Also with a personel from Bharat Digital. Both seem to be conservative, about potential to help Dramas of Discrimination.

These elevator pitches and Australian grant too did not seem good enough as a credibility mark. Or that was my reading of it. But I am already in the audi for the next session that is scheduled to start in a minute.

GiGi Koneti's session on the Quantization of the KV Cache Eviction is one of the rare talks in a tech conference that is also mathetical and also hence appear original enough work.

The hall for this talk is highly packed with a highly technical energy in the talk.

Back to Hall for Ashita's talk on first contribution to leadership in Open Source, Lokesh just finished his talk.

Ashita presents her journey and surveys the pr merge, hacktoberfest etc.

Curiousity is what Ashita labels as the motivation. Being interested and passionate about the mission of the project. And the people, ofcourse. These are the questions Ashita asks and cares.

Ashita suggests taking a small bite and a feature to understand a large codebase. Charts a typical path to contribute: user, reproduce, debug, test first, what+why, discuss.

Industry led projects play bit differently there is CLAs and governance. Open source gives you a diverse feedback even and especially in case of closed PR.

Code isn't the only entry point. Documentation, Support, Triaging the issues on community issues.

Ashita's pain point was trust in the the flutter package, and hence she create Gems to visualise that. Also, contributed couple of PRs.

Maintainers mind set is an evolved one from the contributors and inspire and onboard more contributors.

After the maintainers step, bulidng system and taking the people's game to the next level takes you to the leadership step.

Ashita's take away is that do not treat open source as check box, compititive exam. It is a journey. No short term perks, but long terms are fruitful and it takes time to build trust.

In the age of AI, the trust factor is even more important. You may use AI, but add validation evidences to help the maintainers help your case.

Community calls are a good place to get started and get noticed. The community channels and participation there helps and adds value as well.

Next one is about NHAI workflow streamlining by Deep and Mayank. There is a good context explained about the scale and complexity of the distribution of the issues.

They have an application with MPC server that collects the specific data on this and can be served through this application.

It appears that the talk is actually about the problem and solution the tech dealt with but very little about the tech itself.

I can step out to join the people's digital infrastructure panel actually. Skipping the behavioural chnages this project brought.

This is a very wide and diverse panel with some almost ten people on the stage.

We have Rasam Narsing from Telangana, where access to the fertilizer is lost because of the digitalization of the procurement process.

The common theme, including NREGA wokers, SIR, agree stack issues and other similar issues, people first digital intervention is was absent and created these issues.

I shall now rush to another session on LLVM talk on GPU vs CPUs skipping the next OpenSpeak talk here in Hall 1.

Bhuvan now on GPU vs CPU compalisation. He starts with a simple if statement. CPU takes it as it. But GPU threads it.

I shall now run back to catch the OpenSpeak end. Today, I realised, my notes are a lot less about take aways from these sessions and more about schedule and details. Keeps my learning leaky and limited.

OpenSpeaks is apparently subtitling instrument. Oral Knowledge Framework is getting referred here

It is a nice application. And good talk. But need to now jump to for Physics at 60fps, which is actually about frontend performance.

Deepak is still talking here about the offline mobile apps on Frappe. And showing a video demo of the app without the network.

There was an interesting case where offline app has old schema data, but the now synced online one give is updated schema. Deepak answers it as fail first and submit again approach.

Kumar Anirudha beings his talk with building 3D world. But it needs Inity and Real, and stresses your machine. But browser based engines it's challenging.

Rapier, Rust are being used to build 3D world in browser. And compute is real time in the game engine.

A ground layer and then coordinates from there for the world for interactions. The world rules that define the physics too define that.

The physics engine does what is touching what, where things should push and where does everything end up?

Also forces you feel, including gravity, bounce, surface friction, tension etc get in. These can defined by certain constants at the beginning.

The frame rate also define the allocation you may have for the calculations. And might feel the stuttle. That's where Rust + Wasm come into picture.

Runek is a shadcn for the 3D world. Gets a reference and invites for the contributions. Runek allows you to create a composite component that can be traversed efficiently.

This talk appears to be about Runek and not about the performance itself. These components are static and don't need server.

Kumar also shares a bug where the ground references are fully resolved for the objects on the uneven ground, makig them sinked in the ground.

It's time to move to Hall 1 for Tarunima's talk on NCII on open weight models.

What can we do about non consensual imageries. Both the origin and use of deepfake is predominantly 90% is pornographic. A lot of models doing this on hugging face already.

This is already quite prevalent. AI does enable real world harm. The problem is not sexual content. But it's about non consensual content.

This is only knwoable and probeable on Open Weight models. The conundrum is about how to uphold open weight models while mitigating this.

There are not enough incentives and they are anyway post incident. And also, if your remove nudity from training dataset the overall performance declines. Poisoning the dataset is also a path but harder.

Model developers can add safety evaluation and guardrails. But with the open weight models, these guardrails can be removed.

Why is that hugging face anyways allows these models on their platform. There is possibility that remove the discoverability of the model.

Take ways that are open weight models are big contributors, better mitigation is upstream or downstream, the advocates need to invested intervention elsewhere in the stack. If these models were also open data models, that could have helped us reverge engineer this better

Now the digital infra for the public participation in the archive which are generally not always physically accessible to people. But the Digital spaces are definitely change that possibility.

These two lighting talks have a very different pace and sense of time.

Webpage annotations liek Mosiac and defunct google sidewiki gets a reference. And then Ojas makes a case for the Anno.milli.

Manas then from Tattle takes over to explain the technical architecture. Manas walks us through the UI and Product.

The talk ended. Rahul is here to tell about Foss United activities now. Essentially the annual report.

Foss United Foundation is section 8 without the FCRA. Visions to shape the FOSS and Digital commons. Mission is to support, incubate with grants and public policy changes.

There are campaign programs, learners programs and maintainers program.

MaintainerMay and Forklore to highlight people across, LicencetoLearn to Indian academia on closed sourced software, Local Language FOSS advocacy.

Rahul then walks us through the financials of the Foss United Foundation. Some of the key figures are that the foundation has annual buget of 3cr, 30% of that goes into salaries and 30% of salaries goes to Rahul.

The India Foss takes around 50L to organize and some 15L of that is contributed by ticket sales. Samagata gives 1cr and Zerodha and Frappe give the another. And rest come from other sponsors.

I think, for me, there could be a good support system that could emerge out of this community.

Budhish is one of the governing board, stresses on inspiring individuals and building communities. And hands over to Swastik.

Swastik says that he randomly applied and got into governing board, who was a student then. And he was someone who had a lot of conflicting opinions. He attributed that to the experience difference.

Pourna explicitly mentions reason to join India Foss for diversity, but not just for gender but as well geolocation.