Recommended expert
Murad Ali
AI Agents Automation - LLM-Powered Agentic System
Experience
Jan 2025 - Dec 2025
1 yearAI Agents Automation - LLM-Powered Agentic System
- Developed a multi-agent system connecting LangChain ZeroShotAgent with custom tools for live APIs and task automation.
- Built a FastAPI backend for Jira ticket creation, triage and assignment, auto classification of severity, deduplication, SLA setup, on-call rotation, bidirectional sync of status and comments.
- Added Slack alerts and RAG knowledge lookup with FAISS or pgvector to suggest fixes, optional PagerDuty escalation on policy breaches.
- Orchestrated agents with a router and a Celery plus Redis queue, retries with backoff, rate limits, idempotency keys, human in the loop approvals.
- Implemented guardrails and observability, prompt versioning, token and cost budgets, PII redaction, tool-use allowlists, timeouts, OpenTelemetry tracing, dashboards for accuracy and latency, deployed on Kubernetes with feature flags and canary rollouts.
Jan 2025 - Dec 2025
1 yearShift AI - LLM-Powered Shift Scheduling Assistant
- Built an AI scheduling assistant that automatically generated conflict-free shift plans using LangChain and Llama.
- Designed semantic search with FAISS to interpret constraints and linked schedules with Prisma/PostgreSQL for accurate user data management.
- Delivered a FastAPI backend that simplified predictive scheduling for operations teams.
Sep 2024 - May 2025
9 monthsParis, France
AI Engineer
Eclevar Medtech
- Built a clinical consultation assistant on GCP (Vertex AI and Cloud Run) that surfaces guideline-backed answers during visits. Pilots cut lookup time from minutes to seconds and reduced post-visit documentation by about 25–35%.
- Productionized medical ASR by fine-tuning Whisper and Wav2Vec2 on de-identified audio with VAD and domain lexicons. Word error rate on noisy clinic audio improved by 6 to 8 percentage points and real-time latency became stable.
- Added LLMs (Gemini, GPT, Claude, Llama) for SOAP-style summaries, medication and allergy extraction and risk flags. Used RAG over a vetted corpus to keep answers grounded with citations.
- Used structured reasoning prompts such as chain of thought (COT) and step-by-step decomposition during development and evaluation to improve clinical reasoning and extraction accuracy. In production, kept rationales concise and hid scratchpads to protect privacy and maintain low latency.
- Implemented guardrails including prompt checks, citation requirements, PHI scrubbing and refusal policies. Logged prompts and completions with PII hashing to meet GDPR requirements.
- Set up evals and observability using RAGAS, task-specific EM/F1 and a lightweight human-rating UI. Prevented regressions and shipped safe A/Bs. Median latency dropped by about 30% using quantized inference with vLLM and response streaming.
- Moved prototypes to production with containers, CI/CD, canary releases and drift monitoring. Documented data flows and DPIA materials for compliance.
Mar 2024 - Sep 2024
7 monthsGermany
Master Thesis - Research in Biomedical NLP
Friedrich-Schiller-Universität Jena
- Conducted applied research on Retrieval-Augmented Generation (RAG) for biomedical question answering, working with over 100 scientific papers as source material.
- Built and optimized dense retrieval pipelines with FAISS and improved precision and recall in complex biomedical text comprehension.
- Designed evaluation workflows with RAGAS and achieved strong results in the faithfulness and contextual precision of the generated answers.
Jan 2024 - Dec 2024
1 yearAI-Powered Early Detection of Dyslexia
- Designed an end-to-end AI pipeline for early detection of dyslexia, combining handwriting analysis (Gemini 1.5 Pro) and speech evaluation via ASR.
- Implemented cognitive memory and reading tasks, capturing multimodal behavioral features for prediction.
- Trained a CatBoost classifier to predict dyslexia likelihood with strong accuracy and interpretability.
- Delivered a production-ready pipeline using Python, LangChain and FastAPI, supporting real-time screening use cases.
Mar 2023 - Aug 2024
1 year 6 monthsGermany
AI Engineer (NLP/LLMs)
Incowia GmbH
- Delivered an invoice extraction pipeline that turns scanned PDFs into normalized line-item records using OCR, LayoutLMv3 and an LLM fallback for outliers to lower human review and speed posting.
- Fine-tuned BERT-style NER for vendors, addresses, VAT, IBAN and totals. Achieved over 90% F1 on a versioned golden set and reduced formatting errors with rule-assisted post-processing.
- Extracted tables and line items with structure-aware models and confidence gating. Used Llama or Mistral as a fallback parser for difficult multi-page invoices, improving recall without a spike in false positives.
- Cut cost and latency with dynamic batching, mixed precision, safe 4-bit quantization and document-level caching. Maintained stable p95 latency under load and lowered GPU hours.
- Formalized data and evaluation standards with clear annotation guidelines, inter-annotator agreement checks and CI tests on EM/F1 to block quality regressions.
- Partnered with product and operations to triage failure modes such as skewed scans, stamp overlays and partial tables. Fed fixes back into training and heuristics for steady quality gains.
Jan 2022 - Feb 2023
1 year 2 monthsGermany
Deep Learning Engineer (HIWI)
Max Planck Institute
- Designed an object detection pipeline with SAM and GroundingDINO, improving the classification of biological samples with over 90% precision.
- Enhanced plant species recognition using ResNet-50, raising performance from 85% to 93% and accelerating experimental workflows.
Summary
AI Engineer specializing in Large Language Models (LLMs) and Conversational AI. Experienced in building AI driven self-service systems, automating customer facing workflows and deploying production grade NLP solutions on cloud platforms. Skilled at bridging technology and business outcomes with a track record of improving efficiency, reducing manual work and enhancing user experiences.
Skills
- Programming: Python, C++, C
- Llms & Reasoning: Transformers, Hugging Face, Gemini / Gpt / Claude / Llama, Langchain / Langgraph / Llamaindex, Structured Reasoning Prompts (Chain Of Thought During Development), Tool Calling
- Retrieval & Vectors: Faiss, Pgvector, Hybrid Bm25 + Dense Retrieval, Rerankers, Schema-aware Chunking
- Serving & Mlops: Fastapi, Docker, Ci/cd, Vllm, Quantization, Dynamic Batching, Caching, Streaming, Observability, Ragas, Golden Sets, Em/f1, A/b Testing, Drift Monitoring
- Speech & Document Ai: Whisper, Wav2vec2, Vad, Tesseract, Doctr, Layoutlmv3, Donut, Table And Line-item Extraction
- Cloud & Data: Gcp / Vertex Ai, Sql, Postgresql, Rest Apis
- Ways Of Working: Git, Jira, Confluence, Agile/scrum
Languages
English
AdvancedGerman
ElementaryEducation
Oct 2020 - Jun 2024
Technische Universität Ilmenau
M.Sc. Research in Computer and Systems Engineering · Research in Computer and Systems Engineering · Ilmenau, Germany
Oct 2014 - Jun 2018
UET Peshawar
B.Sc. Computer Systems Engineering · Computer Systems Engineering · Peshawar, Pakistan
Need a freelancer? Find your match in seconds.
Try FRATCH GPT More actions
Similar Freelancers
Discover other experts with similar qualifications and experience