Monthly Archives: November 2025

Why successful people should be mentored?

Leave a reply

Not knowing whether you need a mentor or not at the beginning.

People often assume that mentorship is something you outgrow. They imagine it as a ladder. It’s something you climb early in your career. Once you reach a certain level of success, you step off and stand on your own.

In reality, the opposite is true. The higher you go, the more critical mentorship becomes. I’ve learned this repeatedly throughout my career in open source, in leadership, and in life.

Success Doesn’t Eliminate Blind Spots

When you achieve success, you start to hear less honest feedback. People around you become careful with their words. Colleagues hesitate to challenge your ideas. Slowly, your perspective narrows. It happens not out of arrogance. It’s hard to see what no one reflects back at you.

That’s where a mentor makes all the difference. A good mentor isn’t impressed by your title or your achievements. They see you as the person behind the professional identity. They’ll challenge your assumptions and remind you that growth never stops, no matter how far you’ve come.

Learning Never Ends

The world around us moves too fast for anyone to claim mastery. Technologies evolve, leadership philosophies change, and the definition of success itself shifts over time.

Mentorship keeps you learning. It introduces you to new ways of thinking, new perspectives, and new generations. It forces you to stay curious, and curiosity is what keeps leaders relevant.

In my years working with global database communities, I’ve seen brilliant engineers become stagnant simply because they stopped seeking input. The best ones? They’re still asking questions, still open to being mentored.

Every Step Forward Is New Territory

No matter how experienced you are, every stage of career growth is unexplored terrain. Each new role, responsibility, or challenge introduces conditions you’ve never faced. There are new dynamics, new expectations, and sometimes, new vulnerabilities.

Mentors are the ones who’ve walked those paths already. They know where the turns are, where you stumble, and how to prepare for what’s coming next. They help you see beyond the horizon of your current comfort zone.

That foresight is the ability to anticipate the next chapter of your journey. It is one of the most valuable gifts mentorship offers. It opens your mind to possibilities you have never considered. It helps you approach the unknown with clarity rather than fear.

The Lonely Space at the Top

Leadership is often described as empowering, and it is, but it’s also lonely. You carry responsibilities that few others truly understand. You can’t always be vulnerable with your team or share the full weight of the decisions you make.

Having a mentor gives you a space to breathe. Someone who listens without judging, who helps you find balance when everything feels heavy. Sometimes, mentorship isn’t about advice at all. It’s about presence and perspective. It’s about being reminded that you’re not alone in figuring it out.

From Achievement to Legacy

There’s a point where success stops being about how much you achieve and starts being about what you enable. Mentorship helps you make that shift.

It turns experience into impact. It teaches you how to guide others. It shows you how to pass on lessons without ego. You learn how to translate hard-earned wisdom into something that outlives your career. Every time I’ve been mentored, I’ve become a better mentor myself. I think that is the real cycle of growth.

The Real Value of Mentorship

I’ve come to see mentorship not as a career stage, but as a lifelong relationship with learning. It keeps you honest. It keeps you grounded. And it ensures that success doesn’t harden into comfort.

If anything, mentorship is a mirror. It helps you stay true to your principles. It also connects you to your evolution and your humanity.

No matter how much experience I gain, I’ll always seek mentors. Because the moment I stop learning from others is the moment I stop growing.

This is why I still believe in mentorship even after a successful career. In conclusion, have I had a dedicated mentor in my career? The short answer is no, but I’ve had role models along the way. I’ve used them as my mentors and always asked them what would I do if I were them.

Book Recommendation: https://a.co/d/hc6f6le

Scoped Vector Search with the MyVector Plugin for MySQL – Part II

Leave a reply

Subtitle: Schema design, embedding workflows, hybrid search, and performance tradeoffs explained.

Quick Recap from Part 1

In Part 1, we introduced the MyVector plugin — a native extension that brings vector embeddings and HNSW-based approximate nearest neighbor (ANN) search into MySQL. We covered how MyVector supports scoped queries (e.g., WHERE user_id = X) to ensure that semantic search remains relevant, performant, and secure in real-world multi-tenant applications.

Now in Part 2, we move from concept to implementation:

  • How to store and index embeddings
  • How to design embedding workflows
  • How hybrid (vector + keyword) search works
  • How HNSW compares to brute-force search
  • How to tune for performance at scale

1. Schema Design for Vector Search

The first step is designing tables that support both structured and semantic data.

A typical schema looks like:

CREATE TABLE documents (
    id BIGINT PRIMARY KEY,
    user_id INT NOT NULL,
    title TEXT,
    body TEXT,
    embedding VECTOR(384),
    INDEX(embedding) VECTOR
);

Design tips:

  • Use VECTOR(n) to store dense embeddings (e.g., 384-dim for MiniLM).
  • Always combine vector queries with SQL filtering (WHERE user_id = …, category = …) to scope the search space.
  • Use TEXT or JSON fields for hybrid or metadata-driven filtering.
  • Consider separating raw text from embedding storage for cleaner pipelines.

2. Embedding Pipelines: Where and When to Embed

MyVector doesn’t generate embeddings — it stores and indexes them. You’ll need to decide how embeddings are generated and updated:

a. Offline (batch) embedding

  • Run scheduled jobs (e.g., nightly) to embed new rows.
  • Suitable for static content (documents, articles).
  • Can be run using Python + HuggingFace, OpenAI, etc.
# Python example
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("all-MiniLM-L6-v2")
vectors = model.encode(["Your text goes here"])

b. Write-time embedding

  • Embed text when inserted via your application.
  • Ensures embeddings are available immediately.
  • Good for chat apps, support tickets, and notes.

c. Query-time embedding

  • Used for user search input only.
  • Transforms search terms into vectors (not stored).
  • Passed into queries like:
ORDER BY L2_DISTANCE(embedding, '[query_vector]') ASC

3. Hybrid Search: Combine Text and Semantics

Most real-world search stacks benefit from combining keyword and vector search. MyVector enables this inside a single query:

SELECT id, title
FROM documents
WHERE MATCH(title, body) AGAINST('project deadline')
  AND user_id = 42
ORDER BY L2_DISTANCE(embedding, EMBED('deadline next week')) ASC
LIMIT 5;

This lets you:

  • Narrow results using lexical filters
  • Re-rank them semantically
  • All in MySQL — no sync to external vector DBs

This hybrid model is ideal for support systems, chatbots, documentation search, and QA systems.

4. Brute-Force vs. HNSW Indexing in MyVector

When it comes to similarity search, how you search impacts how fast you scale.

Brute-force search

  • Compares the query against every row
  • Guarantees exact results (100% recall)
  • Simple but slow for >10K rows
SELECT id
FROM documents
ORDER BY COSINE_DISTANCE(embedding, '[query_vector]') ASC
LIMIT 5;

HNSW: Hierarchical Navigable Small World

  • Graph-based ANN algorithm used by MyVector
  • Fast and memory-efficient
  • High recall (~90–99%) with tunable parameters (ef_search, M)
CREATE INDEX idx_vec ON documents(embedding) VECTOR
  COMMENT='{"HNSW_M": 32, "HNSW_EF_CONSTRUCTION": 200}';

Comparison

FeatureBrute ForceHNSW (MyVector)
Recall✅ 100%🔁 ~90–99%
Latency (1M rows)❌ 100–800ms+✅ ~5–20ms
Indexing❌ None✅ Required
Filtering Support✅ Yes✅ Yes
Ideal Use CaseSmall datasetsProduction search

5. Scoped Search as a Security Boundary

Because MyVector supports native SQL filtering, you can enforce access boundaries without separate vector security layers.

Patterns:

  • WHERE user_id = ? → personal search
  • WHERE org_id = ? → tenant isolation
  • Use views or stored procedures to enforce access policies

You don’t need to bolt access control onto your search engine — MySQL already knows your users.

6. HNSW Tuning for Performance

MyVector lets you tune index behavior at build or runtime:

ParamPurposeEffect
MGraph connectivityHigher = more accuracy + RAM
ef_searchTraversal breadth during queriesHigher = better recall, more latency
ef_constructionIndex quality at build timeAffects accuracy and build cost

Example:

ALTER INDEX idx_vec SET HNSW_M = 32, HNSW_EF_SEARCH = 100;

You can also control ef_search per session or per query soon (planned feature).

TL;DR: Production Patterns with MyVector

  • Use VECTOR(n) columns and HNSW indexing for fast ANN search
  • Embed externally using HuggingFace, OpenAI, Cohere, etc.
  • Combine text filtering + vector ranking for hybrid search
  • Use SQL filtering to scope vector search for performance and privacy
  • Tune ef_search and M to control latency vs. accuracy

Coming Up in Part 3

In Part 3, we’ll explore real-world implementations:

  • Semantic search
  • Real-time document recall
  • Chat message memory + re-ranking
  • Integrating MyVector into RAG and AI workflows

We’ll also show query plans and explain fallbacks when HNSW is disabled or brute-force is needed.