Nisar Ahmad

Applied AI Engineer specializing in Agentic AI and Production-Grade RAG Systems

Based in: Lahore, PakistanCurrently: AI Engineer, Excels Tech Solution LLCExperience: 3 years

Hire me

Applied AI Engineer with over 3 years of experience building end-to-end production systems using LLMs, multi-agent workflows, and scalable FastAPI services. He specializes in transforming complex AI research into reliable, containerized solutions deployed on cloud infrastructure like GCP.

Experience

AI Engineer

Excels Tech Solution LLCLahore, PakistanworkDec 2025 – Present

Summary

Architect and build production AI SaaS systems for document intelligence and contextual business applications using LLMs, RAG, and Agentic AI. Design end-to-end RAG pipelines covering document ingestion, chunking, embeddings, vector indexing, hybrid retrieval, prompt orchestration, and LLM response generation. Develop multi-agent AI workflows with agent coordination, contextual retrieval, and task-specific behavior for complex business processes. Own solution architecture, AI workflow design, infrastructure planning, API cost estimation, and retrieval optimization for production deployments. Ship scalable Python/FastAPI services as containerized applications on GCP (Cloud Run, Cloud Build) with CI/CD pipelines.

What I did

  • Architected end-to-end RAG pipelines and multi-agent workflows using LangGraph for production environments.
  • Owned the full lifecycle of production AI SaaS systems for document intelligence, from solution architecture to containerized deployment.

Results

  • Reduced hallucinations by approximately 35% on internal evaluations by improving chunking strategy and implementing a reranking step.
  • Achieved sub-second retrieval latency to provide a real-time experience for end users.
LLMsRAGAgentic AIPythonFastAPIGCPCloud RunCloud BuildCI/CDSolution Architecture

Junior Lecturer

Superior UniversityworkFeb 2025 – Jan 2026

Machine Learning Engineer

PerceptronAILahore, Pakistan · RemoteworkAug 2023 – Aug 2025

Summary

Designed, trained, and deployed ML and deep learning solutions across computer vision, NLP, image classification, object detection, regression, embeddings, and semantic search. Built a facial attractiveness prediction system using teacher-student knowledge distillation to optimize inference performance while preserving model quality. Fine-tuned deep learning models for large-scale image regression and served them through production-ready FastAPI inference services. Designed an automated MLOps training platform that evaluates incoming production data and decides whether model retraining should be triggered or skipped. Implemented automated training, validation, model comparison, versioning, checkpoint management, and promotion of better-performing models. Built computer vision solutions with ResNet and YOLO, and NLP systems for transformer-based classification, embeddings, and similarity search.

What I did

  • Trained a lightweight MobileNet-style CNN student model on a ResNet-based teacher's soft targets and ground-truth labels.
  • Trained and served ML models behind FastAPI inference services.
  • Built an MLOps platform that automatically triggers retraining based on incoming production data.
Computer VisionNLPFastAPIMLOpsResNetYOLOTransformersKnowledge DistillationMobileNet

Skills

Technical

Generative AI
LLMs
FastAPIFastAPI
PythonPython
Advanced RAG
Hugging Face
Hybrid Search (BM25 + Dense)
Prompt Engineering
PyTorchPyTorch
LangGraph
REST APIs
Fine-tuning (LoRA / QLoRA)
Computer Vision (ResNet, YOLO)
Function Calling & Tool Use
Multi-Agent Orchestration
LangChain
NLP
LlamaIndex
Langfuse
DockerDocker
MongoDBMongoDB
Context Management
Reranking
Query Rewriting
Parent-Child Chunking
Qdrant
Milvus
Pinecone
pgvector
CrewAI
Microsoft AutoGen
vLLM
Ollama
Quantization (AWQ, GPTQ, GGUF)
Unsloth
Ragas
DeepEval
Faithfulness & Relevance Metrics
CI/CD
GCP (Cloud Run, Cloud Build)GCP (Cloud Run, Cloud Build)
TypeScriptTypeScript
Node.jsNode.js
NestJS
PostgreSQLPostgreSQL
SupabaseSupabase
TensorFlowTensorFlow
scikit-learn
OpenCVOpenCV
Knowledge Distillation
Llama Guard
TGI
PII Detection
Semantic Kernel
RedisRedis
NeMo GuardrailsNeMo Guardrails
KubernetesKubernetes
SGLang
Helicone
Arize Phoenix
Full-lifecycle product development
Language routing
Barge-in handling
MobileNet

Projects

VocalIQ - AI Voice Customer Experience Platform

Lead Developer / Architectpersonal
Built a full-stack AI voice platform combining real-time voice interaction, RAG, multi-agent workflows, and bilingual Urdu/English speech processing. Engineered browser-based voice interaction with Next.js, Web Audio API, WebRTC, voice activity detection, conversation state management, and barge-in handling. Architected the AI layer on Python/FastAPI with PostgreSQL/pgvector for knowledge retrieval, and TypeScript/NestJS for authentication, workspaces, and multi-tenant infrastructure. Backend AI services (Graph RAG orchestrator, real-time voice pipeline) held as proprietary IP - frontend client open-source; live demo available on request.
Next.jsWeb Audio APIWebRTCPythonFastAPIPostgreSQLpgvectorTypeScriptNestJSGraph RAG
View the code

AI Support Desk

personal
An open-source AI product featuring auto-triage for category, priority, and sentiment, plus drafted replies on a live dashboard.
AI

LLM Gateway

personal
A unified API over LLM providers featuring authentication, rate limiting, streaming, and a cost/latency observability dashboard.
LLM

Education

B.S. Computer Science

COMSATS University IslamabadComputer Science2019 – 2023

Recognition

Certifications

Machine Learning

DeepLearning.AI2025

Advanced Learning Algorithms

DeepLearning.AI2025

Supervised Machine Learning: Regression and Classification

DeepLearning.AI2025
Hire NisarReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis