Musab Javed

Musab Javed

AI Engineer specializing in Agent Systems, RAG Pipelines, and Local LLM Orchestration.

Based in: Lahore, PakistanCurrently: Independent AI Automation Consultant, Independent AI Automation ConsultantExperience: 2 years

Hire me

Musaab Javed is an AI Engineer with over 2 years of experience building production-grade LLM systems and multi-step agentic workflows. He specializes in Python/FastAPI services, structured JSON outputs, and fine-tuning local models for high-performance, cost-effective automation.

Experience

Independent AI Automation Consultant

Independent AI Automation ConsultantRemotefreelanceAug 2026 – Present

Summary

Excelr8 (AI SaaS), Zippit (DTC, 80,000+ customers) and EU/US clients

What I did

  • Shrank campaign launch from hours to minutes (4-8 campaigns/month) with one end-to-end agent workflow
  • Replaced a $1K/month agency with a $50-100/month multi-step LLM pipeline grounded in Pinecone RAG (10-20x cheaper)
  • Developed agentic workflows to automate lead import, enrichment, CRM entry, and reporting.
  • Built a RAG-based SEO content system using Pinecone retrieval and Perplexity citations for Dutch and English health articles.
  • Enforced structured JSON contracts with validation and retries for production workflows.

Results

  • Reduced campaign launch time from hours to minutes by automating end-to-end GTM workflows.
  • Eliminated 14 to 30 hours of manual monthly CRM data entry for a GTM automation platform.
  • Scaled health content production to 2+ articles daily while reducing monthly costs from $1,000 to $50-$100.
  • Achieved 15% to 30% month-over-month organic traffic growth for a Dutch health brand.
  • Cut workflow failures by 60% by enforcing structured JSON contracts with validation and retries.
  • Automated GTM workflows for 4 to 8 campaigns per month.
  • Implemented RAG-based content generation for a brand with over 80,000 customers.
LLMPineconeRAGPerplexityHubSpot

AI Integrations Engineer

TkruptLahore, PakistanworkApr 2025 – Aug 2026
Key Responsibilities: • Reduced analyst review cycles 50% and improved answer consistency 35% with production RAG pipelines (LangChain, Pinecone embeddings) over 5,000+ documents • Shortened integration time 40% with a modular MCP-style tool-use orchestration layer for agents • Lowered failure rates 60% with structured JSON I/O contracts, retries, failure handling and alerting • Saved $15K/year in labour costs across 5+ clients with multi-stage agentic sales automation • Led 3 engineers in scoping and design reviews; primary technical contact for client demos
LangChainPineconeMCPJSONAgentic Automation

AI Engineer

Zaltech AILahore, PakistanworkJul 2024 – Mar 2025

Summary

Key Responsibilities: • Served 1,000+ students (50+ concurrent) at sub-5s with a multi-hierarchical RAG chatbot • Improved retrieval relevance 40% and dropped latency from 3.2s to 1.1s by re-engineering Pinecone + LangChain retrieval • Handled 2K+ daily requests with Python/FastAPI backend services serving OpenAI and Ollama models • Raised domain-task accuracy 40% by fine-tuning Llama 3.1 with Unsloth (LoRA/QLoRA) • Trimmed PostgreSQL query times 55%; gated releases on held-out evaluation sets and regression checks

What I did

  • Scored chatbot changes using custom LLM-as-a-judge prompts to rate relevance and groundedness against expected answers.
  • Utilized RAGAS metrics including context precision and faithfulness for evaluation.
  • Managed DeepEval test cases to track performance across releases.

Results

  • Improved retrieval relevance by 40% by re-engineering the Pinecone and LangChain pipeline.
  • Reduced chatbot latency from 3.2 seconds to 1.1 seconds.
  • Implemented regression checks to ensure new deploys maintained performance against evaluation sets.
  • Built a held-out evaluation set of questions across academic policy, course content, and admin FAQs for Busoga University.
RAGPineconeLangChainPythonFastAPIOpenAIOllamaLlama 3.1UnslothLoRAQLoRAPostgreSQLRAGASDeepEvalLLM-as-a-judge

Skills

Technical

PythonPython
LangChain
LangGraph
Pinecone
FastAPIFastAPI
REST APIs
Ollama
Cursor
PostgreSQLPostgreSQL
LlamaIndex
n8n
MCP
Hugging Face
Unsloth
SQLSQL
pgvector
Letta
vLLM
DockerDocker
FlaskFlask
JavaScriptJavaScript
GCPGCP
AWS
RAG
Agentic Workflows
Constrained Output
DeepEval
LLM-as-a-judge
RAGAS
JSON Schema

Areas of expertise

ISC2 Certified in Cybersecurity (CC)

Projects

Arabic OCR Post-Correction

Local model fine-tunepersonal
Halved word error rate (41.1% to 20.4%) on Arabic OCR documents by LoRA fine-tuning a 0.5B local model on 57k synthetic pairs. Trained in 80 minutes on one laptop GPU; shipped CPU-only on-premise inference with a rewrite guardrail and CER/WER evaluation sets.
Qwen2.5-0.5BLoRAPEFTHugging Face

KrownAI

Client buildprofessional
Built an iOS/Android app where Gemini Flash-Lite turns each hair photo into a structured reading: pattern, density and six scalp signs. Matched routines from 500+ products via pgvector vector search in Supabase Edge Functions at about $0.001 per reading.
ExpoSupabaseGemini Flash-Litepgvector

Whisper Frames

Co-Founder & Founding Engineerpersonal
Owned the backend with co-founders: 500+ assets/week, decode time down 30%, memory 25%; Zindigi Prize 2024 runner-up (85+ startups).
PythonFastAPIPyTorch

Steganogan

personal
A deep learning steganography system fine-tuned on an existing solution using the DIV2K and COCO datasets.
Deep LearningDIV2K datasetCOCO dataset

Education

BS Computer Science

Pakistan Institute of Engineering & Applied Sciences (PIEAS)Computer ScienceSep 2020 – Jun 2024

Recognition

Awards

Zindigi Prize 2024 runner-up

Zindigi2024

Runner-up among 85+ startups

Most Unique Project

PIEAS

Named one of the most unique projects from the graduating class.

Certifications

ISC2 Certified in Cybersecurity (CC)

ISC2

LangChain for LLM Application Development

DeepLearning.AI
Hire MusabReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis