Syeda Eman Saleem

AI Engineer

Syeda Eman Saleem — AI Engineer

Syeda Eman Saleem

AI Engineer

AI Engineer specializing in designing, building, and deploying production-grade Generative AI applications, LLM-powered systems, and agentic AI architectures. Experienced in developing scalable RAG pipelines, multi-agent workflows, AI agents, and high-performance AI backend services using FastAPI, Docker, and cloud-native technologies. Skilled in LLM orchestration, prompt engineering, AI automation, model serving, vector databases, and end-to-end deployment of intelligent applications from research and prototyping to production. Passionate about building reliable, scalable, and impactful AI systems that solve real-world problems.

Syeda Eman Saleem profile image
07
Projects
02
Experience
Contact

Projects

Self-Correcting Code Generation Agent

Built an autonomous AI coding agent that iteratively generates, executes, and self-corrects Python code using live runtime error feedback, requiring zero human

Built an autonomous AI coding agent that iteratively generates, executes, and self-corrects Python code using live runtime error feedback, requiring zero human debugging. Engineered a secure sandboxed execution environment with network isolation, resource limits, and timeout enforcement; architected a provider-agnostic design supporting 4+ LLM providers with zero code changes, plus CLI and Gradio interfaces for real-time monitoring. Implemented full CI/CD automation (GitHub Actions, pytest) and deployed to Hugging Face Spaces, demonstrating production-grade engineering under free-tier infrastructure constraints.

LangGraphOpenAI SDKDockerGitHub ActionspytestGradio

Multi-Tenant LLM Gateway

Engineered a production-grade multi-tenant LLM gateway with per-tenant authentication, rate limiting, and token budget enforcement. Built real-time PII masking

Engineered a production-grade multi-tenant LLM gateway with per-tenant authentication, rate limiting, and token budget enforcement. Built real-time PII masking (Presidio) and prompt-injection detection guardrails, including custom streaming-safe redaction for live SSE responses. Designed a semantic caching layer (ChromaDB, sentence-transformers) to eliminate redundant LLM calls, and implemented multi-provider routing (OpenAI, Anthropic, Ollama) with automatic circuit-breaker failover and full OpenTelemetry/Jaeger observability.

FastAPILiteLLMDockerChromaDBPresidioOpenTelemetryJaeger

LLM Fine-Tuning with QLoRA and Unsloth

Fine-tuned large language models using QLoRA and Unsloth with 4-bit quantization and LoRA adapters, enabling efficient training on limited GPU resources (8GB+ V

Fine-tuned large language models using QLoRA and Unsloth with 4-bit quantization and LoRA adapters, enabling efficient training on limited GPU resources (8GB+ VRAM). Built a complete fine-tuning pipeline from dataset preparation through adapter training to inference. Deployed the trained model via an interactive Gradio interface, hosted on Hugging Face Spaces for public demonstration.

Hugging Face TransformersPEFTPyTorchUnslothGradio

Medical Knowledge Assistant

Built a fully local RAG application enabling plain-English Q&A over medical documents with cited source attribution. Engineered an OCR and PDF ingestion pipelin

Built a fully local RAG application enabling plain-English Q&A over medical documents with cited source attribution. Engineered an OCR and PDF ingestion pipeline (pytesseract, pdf2image, pypdf) supporting both scanned and digital documents. Implemented local embedding generation and vector search (sentence-transformers, FAISS) with zero reliance on external APIs, ensuring full data privacy.

FAISSHugging Face TransformersStreamlitpytesseract

Autonomous Knowledge Discovery Engine

Built a multi-agent AI research assistant with nine specialized agents automating web search, PDF analysis, summarization, cross-source comparison, and report g

Built a multi-agent AI research assistant with nine specialized agents automating web search, PDF analysis, summarization, cross-source comparison, and report generation. Designed autonomous source-comparison and citation systems that detect agreement and disagreement across sources and generate fully cited Markdown research reports with visual charts. Implemented persistent local research memory (SQLite) and optional Playwright browser automation for JavaScript-rendered content, running on a zero-cost, open-source stack.

PythonGroq LLMSQLitePlaywright

OmniRAG-X

Built a production-grade, self-improving multi-agent RAG platform featuring hybrid dense/sparse retrieval and an agentic critique-and-reformulate loop with onli

Built a production-grade, self-improving multi-agent RAG platform featuring hybrid dense/sparse retrieval and an agentic critique-and-reformulate loop with online learning from user feedback. Designed short- and long-term memory and multimodal document ingestion (PDF, DOCX, CSV, images), validated by a RAGAS-style automated evaluation harness achieving a 0.898 overall score. Architected a fully provider-agnostic system (Anthropic, OpenAI, local models) with zero-code-change provider switching and complete agent trace explainability.

PythonFastAPIRAGAS

ExamWatch

Built a real-time cheating detection system that tracks head yaw and pitch angles from pose keypoints to classify student attention as focused or suspicious. En

Built a real-time cheating detection system that tracks head yaw and pitch angles from pose keypoints to classify student attention as focused or suspicious. Engineered a geometric angle-based classification pipeline with real-time annotated video output and visual alerting for exam proctoring applications.

YOLO11m PoseOpenCV

Experience

AI Research Assistant

CAS (Software House)

Mar 2026 — Jun 2026 · 4 mos
Full-timeHybrid
  • Designed and developed a Zero-Shot Non-Intrusive Load Monitoring (NILM) framework leveraging a CNN–Mamba Variational Autoencoder (VAE), semantic embeddings, and contrastive learning, achieving 92.1% classification accuracy on seen appliances and 89.93% generalization on previously unseen device categories.
  • Engineered a federated learning pipeline with cross-dataset benchmarking and distributed model optimization, improving anomaly detection efficiency by 42.6% while enhancing scalability, privacy preservation, and real-world deployment robustness.

Automation Engineer

Ainain Pvt Ltd

Jan 2024 — May 2025 · 1 yr 5 mos
Full-timeHybrid
  • Architected scalable workflow automation systems using n8n, REST APIs, webhooks, and cloud-based integrations, enabling seamless orchestration of business processes across multiple platforms.
  • Engineered production-grade automation pipelines for reporting, data synchronization, and operational workflows, improving efficiency, reducing manual intervention, and enhancing process reliability.
  • Integrated third-party APIs and event-driven services to build resilient, maintainable automation solutions with comprehensive error handling, monitoring, and performance optimization.

Skills

Generative AILLM EngineeringAgentic AIRAG SystemsAI AgentsMachine LearningProduction-Scale AI ApplicationsFastAPIDockerCloud-Native TechnologiesLLM OrchestrationPrompt EngineeringAI AutomationModel ServingVector DatabasesEnd-to-End DeploymentMulti-Agent OrchestrationWorkflow DesignCloud-Native AI InfrastructureAWSGCPAzureInference OptimizationAPI DesignPythonTransformers

Education

BS in Artificial Intelligence

The Islamia University of Bahawalpur

Feb 2022 - Feb 2026
Other

Artificial Intelligence Course

CAS (Center for Advanced Solutions)

Aug 2024 - Jan 2026
Other

Contact

Let’s connect. Choose the fastest way to reach me.

Call

Last updated