Google Summer of Code Selection Guide: Lessons from a GSoC 2025 Pick – Parul University!

Most GSoC selection write-ups treat the process as a checklist: pick a project, contribute, write a proposal, hope for selection. This one treats it as a strategic framework. Krishan Yadav‘s path to a 2025 ML4Science selection contains the practical detail that generic advice cannot.

This article is the tactical companion to Krishan Yadav’s broader journey at Parul University . This article fully covers the Kathmandu-to-Parul journey and faculty network; this one focuses on the Google Summer of Code selection process specifically: how he analysed participating organisations, why he prepared for two preferences rather than one, what differentiated his successful ML4Science proposal from his unsuccessful first-choice attempt, and the concrete advice he offers prospective applicants for Google Summer of Code!

Google Summer of Code is an annual programme that pairs student contributors with open-source organisations on funded summer projects. Each organisation accepts proposals from prospective contributors, mentors selected students through the project work, and provides stipend payments tied to milestone completion. The competition is substantial: hundreds of organisations participate, thousands of students apply, and selection is competitive at every step.

• Application limit. Students can apply to up to three organisations per GSoC cycle. Most successful applicants prepare proposals for at least two.
• Project size categories. Organisations typically offer projects in three size tiers (small, medium, large), with corresponding hour commitments and stipend distributions.
• Selection signal. Mentors evaluate proposals heavily, but also evaluate the contribution patterns and communication quality of applicants in the months leading up to formal selection.

Krishan’s most distinctive strategic move was not in his proposal writing. It was in how he chose which organisations to target.

Most applicants pick organisations based on current-year project listings and immediate interest fit. Krishan inverted this: he analysed historical participation data and prioritised organisations that had been consistently part of the GSoC programme over multiple years. The reasoning was structural. Organisations that had participated since the early years of GSoC (he cited the Python organisation as one example, having participated since 2016) signal strong community support, stable mentorship infrastructure, and long-term institutional commitment to the programme. Newer or sporadic organisations may offer interesting projects but carry a higher risk of inconsistent mentorship or community fragmentation.

• Why historical participation matters operationally – Long-term GSoC participation correlates with organisations that have developed effective mentor training, refined proposal evaluation processes, and built communities capable of absorbing and supporting student contributors. The structural advantages compound across the summer programme.
• Why this approach is non-obvious – Most generic GSoC advice emphasises interest fit, technical skill match, and project quality. These all matter, but they assume the organisation has the operational depth to deliver a good summer experience. Historical participation is a proxy for that operational depth that most applicants do not analyse explicitly.

Krishan’s first preference was an organisation he refers to as Alaska, an organisation he selected partly because of his long-standing personal interest in astronomy, science, and technology dating back to childhood. The work the organisation was doing aligned closely with those interests, making it an apparently ideal first choice. He invested significant time in understanding the organisation, contributing to its community, and preparing his proposal.

The outcome reflected a structural constraint rather than a quality problem. The organisation had requested approximately 16 contributor slots for the cycle. Google allocated 8; many proposals, including Krishan’s, could not be accommodated in the reduced allocation despite the preparation work invested. The lesson is not that the strategy was wrong, but that capacity constraints at the organisation level can override even strong individual applications.

• The structural lesson – Strong proposals can fail not because of proposal quality but because of slot constraints beyond the applicant’s control. This is why the multi-preference approach matters.
• The personal lesson – Krishan had been preparing for Alaska as his primary target, but he had also started preparing his second preference (ML4Science) early, in November, several months before formal applications opened. The early preparation for the backup option became the foundation for his eventual selection.

Krishan’s second preference was ML4Science (Machine Learning for Science), an organisation that applies machine learning techniques to scientific research problems. The fit with his developing ML interest was clear, but what made the difference was the preparation pattern.

• Early start – He began preparing for ML4Science in November 2024, well before formal applications opened in March 2025. Most applicants begin organisation-specific preparation closer to the application opening.
• Sustained contributions – He maintained consistent open-source contributions to the organisation’s projects over the preparation period rather than spiking activity right before the application deadline.
• Mentor engagement – He engaged with mentors through structured questions, draft proposal reviews, and clear communication patterns rather than ad hoc contact.
• Proposal depth – He spent approximately 45 days on his ML4Science proposal, carefully studying project requirements and community expectations before writing.
• Selection result – He was selected for the largest project category (350+ hours of contribution work) with a $3,000 USD stipend distributed across mid-term evaluation and final submission milestones.

ML4Science, like many GSoC organisations, offers projects in three size tiers. Understanding the structure helps applicants align effort and expectations.

• Small projects. Approximately 90 hours of contribution work over the GSoC period. Suited to applicants seeking introductory open source experience or balancing GSoC with other commitments.
• Medium projects. Approximately 175 hours of contribution work. Suited to applicants with developing open source experience who want substantive but bounded project scope.
• Large projects. More than 350 hours of contribution work, typically focused on substantial research-adjacent ML engineering. Suited to applicants with demonstrated technical depth and capacity for sustained engagement. This is Krishan’s category, and that’s why studying B.Tech CSE from Parul University helped him gain his dream exposure.

The project size tier affects stipend distribution proportionally, the depth of mentorship engagement expected, and the contribution intensity required across the GSoC period. Applicants should select tiers based on realistic capacity rather than maximising for stipend alone.

Krishan has been explicit about what matters most in the GSoC selection process. Even strong project ideas get rejected when proposals are not written well. The proposal is the primary artefact mentors evaluate.

• Time investment – Krishan spent approximately 45 days on his ML4Science proposal alone. The depth of preparation is not optional; it is structural.
• Requirements study – He carefully studied project requirements and community expectations before writing, rather than drafting based on general assumptions.
• Mentor feedback loops – He shared each draft with mentors and asked for feedback. Mentors typically have substantial GSoC experience and can identify weaknesses faster than applicants can self-correct.
• Iterative refinement – Multiple draft cycles, each incorporating mentor feedback, produced the final version that was eventually selected.

How applicants communicate with mentors over the months leading up to selection matters as much as the formal proposal. Mentors evaluate communication patterns alongside technical work.

• Submit work before deadlines – If a task is given on Friday, complete it before the next Friday rather than waiting until the deadline. Early completion signals seriousness and consistency.
• Sustained engagement – Many applicants contribute intensively for one week, then disappear for two or three weeks before returning. Mentors interpret these gaps as signals of inconsistent commitment. Regular contributions and discussions build mentor trust over time.
• Communicate proportionally – Krishan’s framing is that applicants should ask questions when genuinely needed rather than disturb mentors unnecessarily. For substantial project work, mentors are usually happy to guide and support. For trivial questions, applicants should attempt self-resolution first.
• Use feedback well – Mentors notice when applicants implement their feedback in subsequent drafts and contributions. Demonstrating that feedback is being absorbed and applied is itself a positive selection signal.

When asked what advice he would give prospective GSoC applicants, Krishan’s answer is not about technical preparation, organisation selection, or proposal writing. It is about getting started in the first place. Many students keep thinking about learning new things, contributing to open source, or participating in programmes like GSoC, but they never actually begin. Once a student takes the first step and starts working consistently, the rest of the journey becomes substantially more manageable. He even said that Parul University provides phenomenal courses such as B.Tech in Artificial Intelligence – ML & Robotics, M.Tech in Computer Engineering, and Master of Science (M.Sc IT) in Artificial Intelligence and many such tech programmes, which ensure dream placements & packages.

Start early. Stay consistent. Keep learning without worrying excessively about the final result. The compounding effect of small daily contributions over months produces the credentials and competence that GSoC selection requires; the absence of starting produces nothing at all.