Machine Learning / AI
Machine Learning Engineer · Local-first AI & semantic search
Muhd Uwais
Machine Learning Engineer building practical, local-first AI systems and semantic search that actually ships.
I solve data-intensive problems by keeping inference local. By replacing cloud-dependency with efficient, on-device pipelines, I deliver systems that are faster, more private, and more reliable than traditional alternatives.
About
Engineering
useful ML systems
I'm building developer tools that work offline-first: semantic code search that understands intent (Ziv), speaker recognition that runs locally (EchoID), and diff viewers that cut through noise (Git Diff, planned).
No cloud dependencies. No API keys. Just systems that ship and solve problems people actually have.
0+
Open-source projects
0+
ML models deployed
0+
Years focused on ML
∞
Lines of curiosity
What I actually build
- Local-first ML tools that respect latency, bandwidth, and privacy constraints.
- Semantic search pipelines combining embeddings, FAISS, and clean APIs.
- Efficient inference systems using ONNX for practical deployment.
Skills
Capabilities across
ML, systems & infra
Backend / Systems
Tools / Infra
Projects
Systems built for
real-world problems
Ziv — Semantic Code Search
Problem: Grepping through large local codebases is slow and misses semantic intent.
- Built a local-first semantic code search tool using FAISS + ONNX embeddings.
- Returns intent-aware results with millisecond latency on commodity hardware.
- Designed feedback loop to improve ranking signal from real usage.
Semantic Git Diff
Problem: git diff compares lines, not meaning — making it impossible to distinguish
a safe rename from a risky logic change without reading every line manually.
- Classifies each change by intent: rename, extract, refactor, format-only, or new logic.
- Scores risk per file using heuristics targeting auth, config, and security-sensitive paths.
- Runs entirely offline with a local ONNX embedding model — no code leaves the machine.
EchoID — Voice Speaker Recognition
Problem: Beginners need a simpler way to build speaker recognition without huge boilerplate or massive datasets.
- Built a binary speaker recognition system using CNNs on mel-spectrograms.
- Added waveform and spectrogram augmentation to improve learning with limited data.
- Supports real-time inference through a GUI-based prediction flow.
Selected Work / Engineering Experience
Learning by
shipping systems
2026 – Present
Independent ML Engineer
Remote · Self-directed build phase
- Designed and shipped Ziv, a local-first semantic code search system for developer workflows.
- Focused on practical retrieval systems, local inference, and tools that solve real usage problems.
- Currently planning Semantic Git Diff, a developer-focused diff review tool built around clarity and speed.
2025 – 2026
ML Projects & System Practice
Brototype · Self-driven execution
- Built EchoID, a speaker recognition project using mel-spectrograms, CNNs, augmentation, and real-time inference.
- Implemented first ML models from scratch, including simple and multiple linear regression using NumPy.
- Built small automation tools and utility projects while strengthening Python, data handling, and engineering discipline.
2024 – 2025
Directed Self-Learning in Python & ML
Brototype
- Followed a direction-based, mentor-light learning model: weekly topics, self-sourced resources, and self-driven practice.
- Built a portfolio website and deepened core Python skills through repeated implementation work.
- Began learning machine learning fundamentals through toy projects, experimentation, and concept-first study.
2022 – 2024
Programming Foundations
Self-directed
- Started with programming basics in C and Java, then transitioned into Python for practical problem solving.
- Built early small programs such as a calculator while learning syntax, logic, and debugging habits.
- Established the base layer of programming discipline that later supported ML and systems work.
What this means in practice
Self-directed execution
Work from goals, find resources independently, and turn unfamiliar topics into working systems without waiting for hand-holding.
Systems-first learning
Move beyond isolated model experiments by building tools, APIs, and local workflows that are usable in practice.
Edge-aware design
Design systems that still work with flaky networks, small machines, and real-world mess.
Visible progression
Show a clear path from programming fundamentals to ML prototypes, then to more serious local-first AI systems and developer tools.
Contact
Work with me on
something concrete
I’m currently open to ML/AI engineering roles focused on practical systems, local-first products, and semantic search. If your team is building useful infrastructure or has an interesting contribution in mind, I’d be glad to hear from you.
Send a Message
Thanks! I'll get back to you soon.