Halil – LLM, RAG, LangChain

A UK-based technology studio that helps organisations turn early-stage AI ideas and ambiguous business requirements into production-ready software. The studio specialises in LLM applications, agentic systems, data automation, and scalable web and mobile platforms.

• Led client engagements from initial discovery and solution definition through architecture, development, and production deployment.
• Worked directly with technical and non-technical stakeholders to identify high-value use cases, challenge unsuitable solution assumptions, and define products that were both technically feasible and commercially worthwhile.
• Designed and delivered production-grade AI and back-end systems using FastAPI, LangChain, LangGraph, cloud infrastructure, and scalable microservice architectures.
• Built multi-agent workflows, RAG applications, and data-driven automation pipelines tailored to specific operational and customer-facing use cases.
• Delivered independent consulting engagements across AI, cloud, and back-end engineering on an Outside IR35 basis.

An AI-powered grant-writing assistant for the funding ecosystem. It helps applicants draft, analyse and strengthen submissions to national and EU funding programmes, using a multi-agent LLM pipeline with a RAG layer that grounds every recommendation in the actual rules of the relevant programme, serving non-technical SME founders and the consultants preparing applications for them.

Halil designed and built the product end to end from the agentic backend and retrieval layer to billing and the freemium go-to-market — and shipped it as a live SaaS, cutting application prep time from days to under an hour.

• Designed and built the full product using FastAPI, React, Redis, Supabase, Stripe, LangChain, and LangGraph.
• Developed a five-step AI analysis engine that evaluated applications across multiple dimensions and generated prioritised recommendations for improvement.
• Implemented a RAG pipeline backed by a vector database to ground recommendations in relevant funding criteria and reduce unsupported AI responses.
• Rebuilt the product through three major iterations based on feedback from early users.
• Launched a product-led freemium model with a paid Pro tier for applicants requiring deeper analysis.
• Reduced grant-application preparation time by approximately 10x—from several days of manual research and drafting to under one hour.

An end-to-end SLA reporting platform for a major Middle East telecom operator, replacing a manual and error-prone process across a large enterprise service portfolio. It let the operator define SLAs as structured rule trees, query metadata, and generate contractual reports automatically serving their service operations and reporting teams.

Halil led the requirements clarification directly with the client’s stakeholders, who described the same SLAs in conflicting ways, and turned that into a formal rule model; then he architected and shipped the FastAPI backend and plugin-based microservices behind it, cutting manual SLA configuration effort by ~40%.

• Led structured discovery sessions with Etisalat’s technical and non-technical stakeholders to resolve conflicting interpretations of SLAs. Translated each definition into an agreed, formally specified rule before committing to the system architecture.
• Designed a flexible SLA rule-tree model capable of supporting varied contractual definitions without requiring a bespoke implementation for every SLA.
• Architected and deployed production-grade REST APIs using FastAPI, providing full CRUD operations for rule trees, metadata queries, and automated report generation.
• Built Python modules that translated declarative SLA rules into generated SQL, with Pydantic-based nested schema validation to identify invalid configurations before report generation.
• Introduced a plugin-based microservices architecture that allowed new SLA types to be added without modifying the core engine, keeping implementation costs manageable as requirements evolved.
• Enabled operations and reporting teams to configure and generate SLA reports more reliably, reducing manual configuration effort by approximately 40%.
• Established a shared, auditable definition of how each SLA was calculated, reducing stakeholder disputes and eliminating configuration errors that had previously surfaced only during report generation.

Built a conversational AI assistant and customer-intelligence models for the sales and marketing teams of a large telecommunications operator. Sales representatives were losing significant time searching for customer and product information scattered across fragmented internal documentation, with no reliable single source of truth.

Halil identified the opportunity, led discovery across sales and operations, built the business case, and presented prototypes and ROI projections to senior leadership before development began. He then designed and delivered a LangChain/RAG-based assistant that allowed field and internal sales teams to ask questions in natural language and receive grounded, contextually relevant answers.

• Led discovery sessions with predominantly non-technical stakeholders across sales, marketing, and operations to identify the workflows where automation would deliver the greatest business value.
• Developed working prototypes and ROI analyses, translating the technical opportunity into a business case that secured executive sponsorship.
• Designed and delivered the generative AI assistant using LangChain, RAG, and FastAPI, taking it from initial prototype to enterprise production within 12 weeks.
• Built customer-profiling and lifetime-value models using XGBoost, Random Forest, KNN, Naive Bayes, and Apriori to improve campaign targeting.
• Added an explainable AI layer so marketing teams could understand why customers were assigned to particular segments or identified as campaign targets.
• Engineered low-latency, highly available API-driven microservices using Django, FastAPI, Flask, GCP, and AWS.
• Achieved a 92% query-resolution rate in production, a 20% improvement in campaign effectiveness, and a 20% operational-efficiency gain across the targeted workflows.

One of Türkiye’s leading telecommunications and technology companies, serving approximately 40 million subscribers at the time and delivering digital services, ICT solutions, and next-generation communication technologies.

Worked as an AI/ML Engineer across computer vision, NLP, OCR, and network automation, developing customer-facing features and internal operational tools.

• Developed a CNN-based virtual-background feature for BiP, a communication application serving approximately two million users, using computer vision to separate users from their surroundings in real time.
• Built an NLP-powered document crawler for the legal team that detected and highlighted key terms across large volumes of legal documents, reducing the effort required for manual review.
• Developed an OCR automation pipeline that extracted structured data from subscriber contracts and stored it in a searchable database, accelerating contract-processing workflows by up to 75%.
• Created an automated monitoring tool to track application and network performance, helping engineering teams detect performance degradation and operational issues.
• Built a Python-based reporting tool that connected to back-end nodes, collected system data, and generated reports to support performance analysis and infrastructure optimisation.
• Collaborated with legal, network operations, and software engineering stakeholders to translate operational requirements into practical AI and automation solutions.

A company that builds digital and ICT solutions, with a focus on innovation, software, and new telecom technologies.

• Performed CNO optimization and automation.
• Used AWS cloud computing and DevOps tools including Ansible, Jenkins, Docker, and Kubernetes.
• Developed programming and scripting solutions with Python.
• Supervised CS and PS services with strong focus on QoS, QoE, and subscriber behavior understanding.
• Analyzed 2G, 3G, and 4G network end-to-end performance and investigated root causes of performance degradations.
• Applied mobile development and testing knowledge using Selenium and Appium.
• Analyzed network KPIs and KQIs from a business perspective.