Skip to content

Commit cd72c82

Browse files
committed
blog: my 2025
Signed-off-by: Chojan Shang <[email protected]>
1 parent 03f5ca3 commit cd72c82

File tree

2 files changed

+209
-0
lines changed

2 files changed

+209
-0
lines changed
Lines changed: 109 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,109 @@
1+
+++
2+
title = "My 2025: The Fifth Year of All-in DB"
3+
description = "A year-end reflection on my fifth all-in year in databases and open source."
4+
date = 2025-12-31
5+
slug = "my-2025-all-in-db-year-5"
6+
7+
[taxonomies]
8+
tags = ["Memoir", "Career"]
9+
10+
[extra]
11+
lang = "en"
12+
+++
13+
14+
If I had not sprained my ankle on Labor Day in 2021, I might not have walked the path
15+
from lakehouse cloud-native to vector multimodal AI native.
16+
Yes, 2025 marks my fifth year of grinding in the database industry.
17+
It is year-end reflection time again.
18+
I recently did a hard retcon of my own history, but the ritual still matters.
19+
This time I want to talk more about myself.
20+
21+
## Just Talking About Work
22+
23+
I still work at a database company.
24+
Yes, "still" means I switched again.
25+
At the end of December, I joined OceanBase to work on open-source projects around
26+
databases and the AI ecosystem.
27+
This is the third database company in my career.
28+
29+
For most of 2025 I worked on GraphRAG at NebulaGraph.
30+
Even though the track is crowded and no longer hot, my team and I still delivered a solid answer
31+
over the past year — 95% accuracy in domain-specific evaluations, and overall cost reduced to a
32+
level comparable to vector plus full-text hybrid indexes.
33+
It almost paved the road for wider adoption of graph-based document RAG.
34+
35+
Work-driven outputs also created some positive impact.
36+
In March, when I researched MCP integration, I designed and implemented the first MCP
37+
integration in the LlamaIndex ecosystem, and contributed it upstream as llama-index-tools-mcp.
38+
After the community refined it further, it became a top 2% download package in the Python ecosystem.
39+
40+
I am not a workaholic or a researcher.
41+
If we talk about code taste, among the colleagues I have worked with, I am probably just average.
42+
Even so, I can still shine.
43+
In the NebulaGraph community awards at the end of 2025, I was recognized by my peers.
44+
Thanks to everyone for the support and help.
45+
Though our paths are far apart, we will meet again.
46+
47+
## May Open Source Bless Us
48+
49+
Because the GenAI platform I work on is a commercial product, my open-source participation
50+
has changed.
51+
More often now, I do things from the angle of requirement research and delivery.
52+
The llama-index-tools-mcp mentioned above is one such result.
53+
54+
I also explored coding agents.
55+
Following ampcode's article, I made a simple Python implementation.
56+
Later explorations became the foundation of my PyCon China 2025 talk,
57+
"Building a Multi-Agent Collaborative CLI Coding Assistant from Scratch."
58+
59+
As a snippet from that talk, I introduced a Python SDK I wrote for the agent client protocol.
60+
I never expected that a few months later it would be integrated by kimi-cli, openhands, and others,
61+
and eventually become part of the official SDK.
62+
Monthly downloads passed one million, and I gained a new role as a maintainer.
63+
64+
I also put work into Apache projects.
65+
I split several OpenDAL subprojects out of the repo, and recently spent some late nights
66+
on the Rust implementation of DataSketches.
67+
68+
Open source might not be better or worse with me in it, but I do draw nourishment and momentum
69+
from it.
70+
The good things I once had seem to have come alongside open source.
71+
Even so, I still have too many promises left behind and failed to deliver.
72+
73+
## Toward a Better Life
74+
75+
The most important change this year is that, because I moved to Hangzhou for work at the end
76+
of the year, I finally ended a four-year long-distance relationship, now in its eighth year.
77+
I also unlocked more memories with my girlfriend, like practicing "goldfish mouth" at G.E.M.'s
78+
concert, and getting fan-made materials from wmls at a Mayday concert.
79+
80+
I tried to pick up new skills through learning.
81+
I can still only play "Lan Hua Cao" by following the sheet, but at least I can recognize
82+
my guitar chords again.
83+
I practiced house dance for a while.
84+
Keeping up with other people's rhythm is already a big challenge for me, but moving with the beat
85+
always helps me relax.
86+
87+
I have two not-so-cute cats.
88+
For a while they would scratch the door open at midnight and jump onto me to rave.
89+
Honestly, I have not taken good enough care of them.
90+
But Goji still shows me his little belly, and Aloe has been willing to come over for a tiny bite
91+
of cat treat after we moved to Hangzhou.
92+
93+
Life, oh life.
94+
Days and nights, but when I pick up the pen I do not know what to write.
95+
Things are okay, no big storms.
96+
In a recent checkup, I did not see fatty liver on the report anymore.
97+
98+
## A Few Lines for AI
99+
100+
Aside from some changes in efficiency, process, and rhythm, my work and life have not been
101+
stirred much by AI.
102+
Value does not shift dramatically just because one or two tools are added.
103+
104+
I do not plan to count how much code I generated with AI, or how many tokens or dollars I spent
105+
in the past year.
106+
But, rarely, this is a piece of writing typed 100% by me.
107+
108+
The world is going through profound change.
109+
I am honored to be part of it, one wave among countless, breaking against a rock.
Lines changed: 100 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,100 @@
1+
+++
2+
title = "我的 2025:All in DB 的第五年"
3+
description = "写在第五个数据库年头的年终回顾:工作、开源与生活。"
4+
date = 2025-12-31
5+
slug = "my-2025-all-in-db-year-5"
6+
7+
[taxonomies]
8+
tags = ["Memoir", "Career"]
9+
10+
[extra]
11+
lang = "zh"
12+
+++
13+
14+
如果 21 年的五一没有崴到脚,我可能也不会一路从湖仓一体云原生走到向量多模 AI Native。
15+
是的,2025 年是我在数据库行业摸爬滚打的第五个年头。
16+
又到写年终总结的时候了,虽然前段时间狠狠地搞了一下「岁月史书」,但是该有的仪式感还是要的。
17+
18+
这次多谈谈自己。
19+
20+
## 只是谈谈工作
21+
22+
我还是在一家数据库公司工作。
23+
是的,「还是」意味着我又一次切换了工作。
24+
12 月底的时候,我加入 OceanBase 从事数据库和 AI 生态相关的开源工作,
25+
这已经是我职业生涯中的第三家数据库公司。
26+
27+
当然,在 NebulaGraph 做 GraphRAG 占据了 2025 的绝大多数时间。
28+
尽管这个赛道已经过度拥挤并且不再火热,但我和团队里的其他同学在过去一年里还是交出了一份不错的答卷——
29+
在特定领域的客户评测中能够达到 95% 的准确率,
30+
总体开销也降低至可以对标向量全文混合索引,
31+
几乎为基于图的文档类 RAG 的广泛应用铺平了道路。
32+
33+
至于工作派生出的产出也产生了一些积极的影响。
34+
比如在 3 月份我调研集成 MCP 的时候,
35+
设计和实现了 LlamaIndex 生态中的第一个 MCP 集成方案,
36+
并且以 llama-index-tools-mcp 的形式贡献给上游。
37+
在由社区进一步开发完善后,它成为了整个 Python 生态中下载量排在前 2% 的包。
38+
39+
我算不上一个醉心工作和研究的人。
40+
若论代码品味,在过往共事过的同事中似乎也只算一般。
41+
即便这样的我,也可以发光发热。
42+
在 2025 年底的 NebulaGraph 社区评选中,也获得同事的肯定。
43+
非常感谢大家的提携和帮忙,尽管江湖路远,终究还是会有机会再见面。
44+
45+
## 愿开源庇佑你我
46+
47+
由于工作从事的 GenAI 平台是一个商业化产品,我参与开源的模式也发生了改变,
48+
更多时候会从需求调研和落地的角度做一些事情,
49+
前述的 llama-index-tools-mcp 就是这样的一个产物。
50+
51+
当然,我围绕 coding agent 做过一些探索。
52+
参照 ampcode 的文章做了一个简单的 Python 实现,
53+
它的后续演进的一些探索成为了在 PyCon China 2025 上的分享
54+
《从 0 开始构建多 Agent 协作的命令行编程助手》的基石。
55+
56+
作为演讲中的一个片段,我介绍了自己为 agent client protocol 做的 Python SDK。
57+
没有想过几个月后它会被 kimi-cli、openhands 等集成,
58+
并且最终成为官方 SDK 的一员,单月下载量过百万,我也多刷新出一个新的维护者的身份。
59+
60+
我在 Apache 项目上也有一些投入。
61+
将 OpenDAL 的几个子项目从 Repo 中拆了出来,
62+
最近又在 DataSketches 的 Rust 实现上投入了一些无处安放的夜晚。
63+
64+
开源是否有我并不见得会更好或更糟糕,但我确确实实会从中汲取养分和动力。
65+
哪些我曾经拥有过的美好一切,似乎就是在和开源伴生而来。
66+
尽管我还有太多承诺过的事情被抛在脑后,没有办法及时兑现。
67+
68+
## 奔向更美好的生活
69+
70+
今年最重要的改变是:由于年底换到杭州工作,我总算结束了长达 4 年的异地恋爱,在第 8 个年头。
71+
也解锁了更多和女朋友的美好的回忆,
72+
比如在邓紫棋的演唱会上练习金鱼嘴,在五月天的演唱会被 wmls 投喂物料。
73+
74+
尝试通过学习来收获一些新技能。
75+
虽然现在也还只是停留看着谱子弹《兰花草》的水平,
76+
但至少又再一次能认得吉他和弦了。
77+
练习了一段时间 house 舞种,
78+
尽管跟上其他人的节奏对我已经是一个非常大的挑战,
79+
但跟着节奏摇摆总是能让心情得到放松。
80+
81+
我有两只不那么可爱的猫咪。
82+
有段时间一到凌晨就会扒拉开门跑到我的身上蹦迪。
83+
坦白说,我没有把它们照顾很好。
84+
但是枸杞还是一如既往地向我展露它的小肚皮,
85+
芦荟在搬家到杭州以后也愿意偶尔凑过来吃一点点猫条。
86+
87+
生活啊,生活呵。
88+
日日夜夜,但是提笔就不知道要写些什么了。
89+
过得还算安好,也无甚风雨。
90+
不久前的体检里似乎也没有再看到脂肪肝的影子。
91+
92+
## 留一些笔墨给 AI
93+
94+
除了一些效率、流程和节奏上的变化,我的工作和生活并没有因为 AI 而产生更多的波澜,
95+
毕竟价值并不会因为多了一两个工具的介入就发生显著的偏移。
96+
97+
我没有打算统计我有多少代码是使用 AI 生成,过去一年到底花费了多少 token 或者费用。
98+
但是,难得地,这是一篇 100% 我手打的文字。
99+
100+
世界在发生着深刻的变革,很荣幸能够参与其中,作为无数浪花中拍在石头上的一朵。

0 commit comments

Comments
 (0)