2025 was the year everything changed.
- Bought my first home.
- Gave my first international conference talk.
- Shipped AI tools used by Fortune 500 companies.
Here's my honest 2025 wrap: the wins, the failures, and everything in between.
The two moments that defined my year
May 2025 - We bought our first home in Jabalpur. After years of remote work at Gitpod (now Ona), open source grind, and GitHub Sponsorships - it finally became real. Not just numbers in an account. A home. Ours.
Big thanks to open source, sponsors, Gitpod, and my parents who made this possible.
August 2025 - I gave my first-ever talk at KubeCon India in Hyderabad. Standing on that stage, speaking to hundreds of cloud-native engineers from around the world... surreal doesn't cover it.
From writing my first PR to speaking at KubeCon. If you told 2020 me this would happen, I'd have laughed.
Work: Gitpod → Ona
The company rebranded. My role evolved. Shifted from cloud dev environments to AI infrastructure. Building agent systems for real enterprise customers.
The highlight? We 4x'd daily AI agent sessions for a Fortune 500 finance company. In a single week. One engineer drove all of it.
That kind of impact is addictive.
The year I went all-in on MCP
Model Context Protocol is the "USB-C for AI" - and I bet big on it.
MCP servers I shipped:
- Dev.to publisher
- Apple Notes integration
- Memory Journal (iCloud photos)
- Python & TypeScript sandboxes
- Claude Code async runner
Each one taught me something new about giving AI agents "hands" to do real work.
Writing: 20+ technical articles
Most impactful:
- The Engineering Guide to Context Window Efficiency
- Securing Agentic AI: Authorization Patterns for Autonomous Systems
- Why Agent Orchestration is Harder Than Kubernetes
- Claude Code is Costly - Unless You Do This
- AWS S3 Vectors at Scale: 10 Million Vectors Benchmarked
Multiple posts hit Dev.to's "Top 7 Must Reads."
Open source: still the foundation
Maintained:
- OpenFGA (Google Zanzibar-inspired auth)
- github1s (23K+ stars!)
395 repos. 958 GitHub followers.
Open source isn't a side thing. It's the thing.
Projects shipped
- Distill - context deduplication for LLMs
- SageMap - thoughts to interactive contradiction maps
- ArchiFusion - AI architectural designer
- MediBrief - research papers to video summaries
- Cloud Architect AI - natural language to infra diagrams
- AgentFlow - agent orchestration engine
Each one = a crash course in LLMs in production.
Research I'm proud of
- Benchmarked AWS S3 Vectors at 10M vectors
- FAISS vs NMSLib vs S3 Vectors comparison
- Context engineering for production LLM systems
- Agent authorization with OpenFGA
- Lambda cold start elimination with LLRT
Turned each into posts with interactive visualizations.
Community growth
- First international conference talk (KubeCon India!)
- Active mentoring on MentorCruise
- GitHub Sponsors live
Building in public works. Not overnight. But steadily.
What I learned
- Context engineering > prompt engineering
- Agent security is the next big challenge (traditional RBAC fails for autonomous systems)
- MCP will become essential infrastructure
- Consistent publishing beats viral moments
- Remote work + open source = real wealth (not just money - freedom, impact, ownership)
What didn't work
- Built too many things at once
- Didn't say "no" enough
- Delayed rest until burnout hit
- Underestimated how long quality content takes
Failures are data points. Nothing more.
2026 focus
- Distill
- More conference speaking
- More research / deep work
- Deeper technical writing (fewer, better)
- Building while staying healthy (non-negotiable)
Final thought
From a tier-3 college in India to:
- Engineering at a global company
- Maintaining 23K+ star repos
- Speaking at KubeCon
- Shipping AI tools for enterprises
- Owning a home
The path was never linear.
But every PR, every blog post, every connection compounds in ways you can't predict.
Consistency isn't boring. It's the cheat code.