Muhammad Awais Arif

AI/ML Engineer | Architecting High-Precision Agents & Real-Time Trading Systems

Based in: Lahore, PakistanCurrently: AI/ML Engineer (FlavorWiki client project), CodeBricks GlobalExperience: 1 year

Hire me

AI/ML Engineer specialized in building high-accuracy agentic pipelines and real-time distributed systems. From reducing LLM prompt overhead by 72% at FlavorWiki to engineering a multi-stack copy-trading ecosystem, I focus on bridging the gap between complex statistical modeling and production-ready software.

Experience

AI/ML Engineer (FlavorWiki client project)

CodeBricks Globalwork2026 – Present

Summary

Integrated PLS regression and statistical analysis into the FlavorWiki agent; built preference mapping across 23 products with Q2 = 0.495 using three components. Built an automated survey-theme and aspect-sentiment pipeline with local lightweight models, achieving 93.7% accuracy in internal evaluation and replacing the data team's manual classification workflow, without external AI APIs or LLMs. Built a 41-rule validation pipeline for 7,500 survey rows across 25 products, integrating data-cleaning workflows into consumer sensory analytics. Integrated MapleRepair and SQLFluff-based SQL validation and repair into the agent workflow, alongside PostgreSQL-aware SQLGlot tooling for identifier, alias, and join errors. Prototyped context compaction that reduced initial agent prompt size by 72.1% while retaining instructions and SQL safeguards; documented the design for integration review.

What I did

  • Built an analytics agent that decomposed complex survey data research questions into controlled steps using a multi-node workflow.
  • Implemented a decision loop for SQL generation that included validation and automated repair steps before execution.
  • Integrated deterministic tools like SQLFluff and MapleRepair into the agentic workflow to handle formatting and join query repairs.
  • Built an NLP pipeline for analyzing open-ended customer responses using spaCy for noun-chunk extraction, lemmatization, and text preprocessing.
  • Generated embeddings and used agglomerative clustering and cosine similarity to discover recurring themes automatically.
  • Performed sentiment analysis at the theme level to identify specific positive or negative feedback points.
  • Integrated PLSR workflows into an agentic analytics system, enabling the agent to decide when to use the model and interpret results.

Results

  • Achieved 93.7% accuracy in internal evaluation for survey-theme and aspect-sentiment pipeline.
  • Reduced initial agent prompt size by 72.1% through context compaction prototyping.
  • Successfully replaced manual classification workflow for the data team.
PLS regressionStatistical analysisLocal lightweight modelsMapleRepairSQLFluffSQLGlotPostgreSQLPythonNLPLangGraphLangChain

Software Engineering Intern

GigalabsLahore, PakistaninternshipAug 2023 – 2024

Summary

Delivered a React, NestJS, and PostgreSQL social platform as sole developer for 50+ employees, covering requirements, APIs, database design, and UI state. Designed normalized TypeORM schemas, role-based access controls, and enterprise validation rules; added Redux Toolkit for application state management.

What I did

  • Built a social platform from scratch for employees to share pictures and descriptions.
  • Planned features for event tracking, gaming room integration, and music rooms for leisure and work hours.

Results

  • Acted as sole developer for a social platform serving 50+ employees.
ReactNestJSPostgreSQLTypeORMRedux Toolkit

Skills

Technical

PythonPython
Next.jsNext.js
Aspect-Based Sentiment
TypeScriptTypeScript
LLM Agents
PostgreSQLPostgreSQL
GitGit
NestJS
spaCy
Embeddings
PLSR
NumPy
ReactReact
pandas
Statistical Modelling
Experiment Design
VercelVercel
LangChain
LangGraph
FastAPIFastAPI
REST
PyTorchPyTorch
GraphQLGraphQL
SQLGlotSQLGlot
SQLFluffSQLFluff
tRPC
gRPC
Agglomerative Clustering
MapleRepair
DockerDocker
WebSockets
CUDA
Parallel and Distributed Computing
Robotics
Visual Programming
.NET
MVC Architecture
React NativeReact Native
ExpressJSExpressJS
Cosine Similarity
Text Preprocessing
LoRA
QLoRA
PEFT
Model Quantization
Fine-tuning Evaluation

Areas of expertise

Requirements Translation

Projects

ReAct vs Native Tool-Calling Agent

Developerpersonal
Built ReAct and native function-calling harnesses over shared tools with parser checks, retries, step limits, and token accounting.
PythonGroqLLM APIs

Resume Forger

Developerpersonal
Shipped an AI resume builder with structured LLM output, PDF generation, ATS checks, and a live Vercel deployment.
Next.jsTypeScriptLLM APIsVercel
Visit the site

TradeSync Pro, MT5 Copy-Trading Ecosystem

Lead Developerprofessional
Delivered a real-time system spanning two Python desktop clients, a NestJS gateway, an MSSQL store, and a Next.js control dashboard. Translated trading requirements into per-account risk controls, symbol mapping, and signal distribution across connected accounts.
Next.jsNestJSPythonMSSQL

Parallel Video Editor

personal
Developed a video editor with various effects that utilized GPU parallel processing and compared performance against CPU core-level execution.
CUDA

Reinforcement Learning Navigation Robot

personal
Built a robot in Webots that used reinforcement learning to learn from mistakes and improve navigation within a custom environment.
Webots

API Latency Testing Application

personal
Built an application to test and compare latencies across REST API, tRPC, and gRPC protocols.
RESTtRPCgRPC

Car Price Prediction

personal
A personal project using real scraped vehicle data to predict car prices.
Python

Music Replication Experiment

personal
An experiment attempting to replicate Fur Elise using a simple MLP model to observe the effects of overfitting and underfitting.
MLP

Education

Bachelor of Science

COMSATS University Islamabad, Lahore CampusComputer Science2026
Hire MuhammadReach out about a role, a contract or a conversation.For recruiters

Muhammad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis