Mishal Ashraf

AI Engineer specializing in Agentic AI and Enterprise Retrieval Systems

Based in: Lahore, PakistanCurrently: Associate AI Engineer, FieldforceExperience: 4 years

Hire me

AI Engineer with extensive experience building LLM applications, multi-agent workflows, and hybrid search retrieval systems. Expert in agent observability and evaluation using Langfuse and LLM-as-judge frameworks to ensure high-quality enterprise AI solutions.

Experience

Associate AI Engineer

FieldforceLahoreworkJun 2025 – Present

Summary

Enterprise Retrieval System • Built a retrieval system over enterprise records (people, suppliers, organizations, and projects), exposed to an AI agent as search and lookup tools. • Implemented hybrid search combining vector embeddings (pgvector) with keyword search, and evaluated retrieval quality on a golden question set. Agent Observability and Evaluation • Integrated Langfuse into a LangGraph tool-calling agent for tracing, prompt management, and observability. • Built an evaluation pipeline combining rule-based checks of agent behavior with an LLM judge for response faithfulness, validated against human labels.

What I did

  • Planned to expand the evaluation dataset using real usage queries to cover more scenarios.
  • Built a dashboard for company employees to view details of organizations, projects, and involved personnel.
  • Maintained cadence groups, ownership, and relationship matches of groups and people.

Results

  • Increased Recall@1 from 92% to 93% and Recall@5 to 98% by fixing matches for names and emails.
  • Increased retrieval latency by 2 milliseconds
  • Evaluated system performance using a golden question set of 65 queries
LLMAI AgentpgvectorVector EmbeddingsHybrid SearchLangfuseLangGraphLLM-as-judgeDashboard development

Online AI/ML Instructor

PreplyRemotepart timeJul 2022 – Present

Summary

• Delivered personalized AI and Machine Learning training to 60+ international students. • Designed hands-on learning paths covering Python, Machine Learning, and Deep Learning.

Results

  • Trained 60+ international students
PythonMachine LearningDeep Learning

Research Student

TUKL Lab, NUSTIslamabadworkNov 2023 – Nov 2024
• Built a multi-agent legal document summarization system using CrewAI. • Developed evaluation pipelines using ROUGE and BERTScore to measure summary quality. • Performed human-centered evaluation to assess the factual consistency of generated summaries.
CrewAIROUGEBERTScoreMulti-agent systems

Skills

Technical

RAG
LLMs
Prompt Engineering
PythonPython
Multi-Agent Systems
pgvector
Vector Embeddings
Semantic Retrieval
Hybrid Search
LLM-as-Judge
Langfuse
Agentic AI
Hugging Face Transformers
Tool-Calling Agents
LangChain
LangGraph
CrewAI
ROUGE
Indexing
BERTScore
Tracing
Prompt Versioning
PostgreSQLPostgreSQL
Hallucination Evaluation
Dashboard development

Projects

Enterprise Retrieval System

Associate AI Engineerprofessional
Built a retrieval system over enterprise records (people, suppliers, organizations, and projects), exposed to an AI agent as search and lookup tools. Implemented hybrid search combining vector embeddings (pgvector) with keyword search, and evaluated retrieval quality on a golden question set.
pgvectorVector EmbeddingsHybrid SearchAI Agent

Multi-agent Legal Document Summarization System

Research Studentacademic
Built a multi-agent legal document summarization system using CrewAI. Developed evaluation pipelines using ROUGE and BERTScore to measure summary quality. Performed human-centered evaluation to assess the factual consistency of generated summaries.
CrewAIROUGEBERTScore

Employee Information Dashboard

personal
A side project deployed for company employees to access information regarding organizations, projects, and personnel, including cadence groups and ownership relationships.

Education

MS in Artificial Intelligence

National University of Sciences and Technology (NUST)Artificial IntelligenceGPA 3.4Sep 2021 – Nov 2024

BSc in Computer Engineering

Bahauddin Zakariya University (BZU)Computer EngineeringGPA 3.1Nov 2015 – Aug 2019

Recognition

Certifications

Machine Learning Specialization

Coursera

Deep Learning Specialization

Coursera

NLP Specialization

Hugging Face

Multi-Agent Systems with CrewAI

DeepLearning.AI

Claude 101

Anthropic Skilljar

Claude Code in Action

Anthropic Skilljar

AI Fluency: Framework & Foundations

Anthropic Skilljar
Hire MishalReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis