Senior AI & LLM Backend Engineer

RESUME MANISH PITHWA Vaibhav Nagar Ext., Kanadiya Road, Indore 452016 (M.P.) INDIA Mob.: Email : LinkedIn: [контакт скрыт] GitHub: [контакт скрыт] OBJECTIVE: To pursue a highly challenging career in the IT field, where I can apply my knowledge, acquire new skills and work with a team of highly experienced professionals. TECHNICAL SKILLS: AI Frameworks/Libraries: PyTorch, Hugging Face Transformers, TensorFlow, LangChain, FastAPI, Gradio, Pydantic, scikit-learn, Pandas, NumPy AI/ML: LLM fine-tuning, Embeddings, Attention Mechanisms, RAG, Model Distillation, Quantization, Pruning, Sentiment Analysis, Text Summarization, Semantic Search, NER, Supervised/Unsupervised Learning APIs/Platforms: OpenAI, Anthropic Claude, Meta Llama, Vertex AI, Google Gemini, Hugging Face, LiveKit, ChromaDB, LlamaIndex, crewAI DevOps: Docker, Kubernetes, Git, CI/CD (CircleCI, GitHub Actions), Ansible, Jenkins, Terraform, Prometheus, Grafana, ElasticSearch, Logstash, Kibana (ELK) Other: Playwright, AsyncIO, UMAP, MLflow, Arize Phoenix Cloud: AWS, GCP Programming : PHP, Python, Node.JS, Javascript, Bash. Backend Frameworks : Laravel 5.4/6/8, Lumen, Zend (Laminas) MVC, CakePHP, Yii. Frontend Frameworks : Angular 2/4/5/6/12, AngularJs, Vue.Js Open Source Packages : Wordpress, Joomla, Dolphin,. Containers: Docker, Kubernetes, DDEV Message Queue: RabbitMQ, AWS SQS Operating System : Ubuntu, Windows, UNIX, LINUX. Back end Tools : MySQL, PostgreSQL, Oracle CERTIFICATIONS ● DeepLearning.AI - Agentic AI and Generative AI Specialization ● DevOps Work Experiences :- Organization: Micronisus Systems, Dubai, UAE Designation: Senior AI and Backend Developer Duration: 4 Months, 1-Dec-2025 to Mar-2026 Job Profile: Results-driven AI/LLM Backend Engineer with expertise in designing and deploying enterprise-grade GenAI solutions using Large Language Models, RAG pipelines, and agentic AI systems. Experienced in building scalable backend architectures, integrating vector databases, and orchestrating multi-agent workflows for complex problem-solving. Strong background in Python-based microservices, cloud-native deployments, and MLOps practices, with a focus on performance optimization, observability, and secure AI system design. Proven ability to lead end-to-end AI initiatives—from prototyping to production—while collaborating with cross-functional teams to deliver impactful, intelligent applications. Technology Stack: AI / LLM / GenAI • LLMs: OpenAI API, Anthropic Claude, LLaMA, Mistral, Falcon • Techniques: RAG (Retrieval-Augmented Generation), Prompt Engineering, Embeddings, Semantic Search • Model Optimization: LoRA, Quantization (GGUF), Knowledge Distillation Agentic AI & Frameworks • Frameworks: LangChain, LlamaIndex • Multi-Agent Systems: LangGraph, CrewAI, AutoGen • Protocols: Model Context Protocol (MCP) Vector Databases & Search • Vector DBs: Pinecone, Weaviate, Milvus, ChromaDB • Libraries: FAISS • Hybrid Search & Indexing Strategies • BaaS: Supabase Backend Development • Languages: Python • Frameworks: FastAPI, Flowable, Openbao, Valkey, • API Design: REST, async services, microservices architecture ️ Cloud & DevOps • Cloud Platforms: Amazon Web Services, Google Cloud Platform, Microsoft Azure • Containerization: Docker • Orchestration: Kubernetes • CI/CD & MLOps pipelines Performance & Observability • Caching Strategies (Redis, in-memory caching) • Monitoring & Logging (Prometheus, Grafana, ELK Stack) • Latency Optimization & Cost Optimization for LLM APIs Multimodal & UI (Optional / Bonus) • Speech AI: Transcription, Text-to-Speech (TTS) • Rapid Prototyping: Streamlit, Gradio • Frontend Integration: React (for AI dashboards/tools) Security & Enterprise Readiness • Secure API Design, Authentication & Authorization • Data Privacy & Compliance (PII handling, encryption) • Scalable, fault-tolerant architecture Organization: First Due Solutions India Pvt. Ltd. Designation : Senior PHP Developer Job Profile: Working on Yii/Vue.JS, DDEV and Docker to build website, Web Services and API integration. Build . Develop frontend in Vue.JS consuming Yii Web Services. Deployment on AWS servers via Github repositories on Ubuntu/Nginx/PostgreSQL. Technology Stack: Yii, Vue.JS, Apache/Nginx, MySQL/PostgreSQL, PHP 7.4, API integration, Docker, RabbitMQ, Redis, AWS, S3 Duration: 1 Year 10 Month 10-Jul-2023 to 16-May-2025 Organization: SmartCloud InfoFusion Pvt. Ltd. Designation: Lead Software Engineer II Job Profile: Working on Laravel/Angular and Docker to build website, Web Services and API integration. Build . Develop frontend in Angular 12 consuming Laravel Web Services. Deployment on Digital Ocean servers via Github repositories on Ubuntu/Nginx/MariaDB. Technology Stack: Laravel, Angular 12, Apache/Nginx, MySQL/MariaDB, PHP 8, API integration Duration: 1 Month 22 days 13-Mar-2023 to 5-May-23. Organization: PinkNBlu Infomedia Pvt. Ltd. (Product Based Company) Designation: Head Of Technology (HOT) Job Profile: Leading the team of 5 resource comprising Laravel/Android developers, designers, Project Manager, Data Analyst and QA. Getting product enquiries from business/marketing team and converting into business deliverables. Requirement gathering, analysis, solution suggestion, estimating cost/time, identifying challenges and required resources, preparing execution plan, allocating resources, managing resources among multiple modules/tasks, execution of projects, updating progress to business/marketing team, monitoring project status and cost, providing feedback to team members, helping them achieve project deliverables under estimated time by working on development stacks, providing testable modules to QA and getting quality assured, deploying product deliverables on servers, communicating with business/marketing team for feedback and follow-up for new business and other stuff required to deliver successfully product to business/marketing team. Technology Stack: Laravel, Lumen, Angular 2/4/5/6, AngularJs, Apache/Nginx, MySQL/MariaDB, PHP 5/7, WordPress, CakePHP, Yii, Zend (Laminas), Joomla, API integration, AWS, EC2, RDS, S3, SQS, SNS, ElasticSearch Duration: 2 Years 7 Months 17-Aug-2020 to 10-Mar-2023. Organization: Intense Technologies Limited (NSE: INTENTECH) Designation: Lead Fullstack Job Profile: Leading the team of 5 resource comprising PHP/Wordpress/CakePHP/Laravel/Angulardevelopers, designers, QA and Business Development Executives. Getting project enquiries from client and converting into business deliverables. Requirement gathering, analysis, solution recommendation, estimating cost/time, identifying challenges and required resources, negotiating client for time/cost, preparing execution plan, allocating resources, managing resources among multiple projects/clients, execution of projects, updating progress to clients, monitoring project status and cost, providing feedback to team members, helping them achieve project deliverables under estimated time/cost by working on development stacks, providing testable modules to QA and getting quality assured, deploying project deliverables on servers, communicating with clients for feedback and follow-up for new business and other stuff required to deliver successfully projects to clients. Technology Stack: Laravel, Lumen, Angular 2/4/5/6, AngularJs, Apache/Nginx, MySQL/MariaDB, PHP 5/7, WordPress, CakePHP, Yii, Zend, Joomla, API integration, CS-Cart, Java Spring Duration: 6 Months 21-Jan-2020 to 14-Aug-2020. Organization: Self Entrepreneur Designation: Project Manager Job Profile: Working on Laravel to build websites, admin, Web Services and API integration. Build Web Services using Lumen. Develop frontend in AngularJS/Angular 2/4/5/6 consuming Laravel/Lumen Web Services. Deployment on Digital Ocean servers via Bitbucket repositories on Ubuntu/Nginx/MariaDB. Technology Stack: Laravel, Lumen, Angular 2/4/5/6, AngularJs, Apache/Nginx, MySQL/MariaDB, PHP 5/7, WordPress, CakePHP, Yii, Zend (Laminas), Joomla, API integration From: 2 Years 6 Months 30-June-2017 to 20-Jan-2020. Organization: Laxyo Solution Soft Pvt. Ltd. (A Laxyo Group of Companies) Designation: Sr. TL (Team Lead) Job Profile: Leading the team of 5-10 resource comprising PHP/Wordpress/CakePHP/Laravel/Angular developers, designers, QA and Business Development Executives. Getting project enquiries from client and converting into business deliverables. Requirement gathering, analysis, solution suggestion/recommendation, estimating cost/time, identifying challenges and required resources, negotiating client for time/cost, preparing execution plan, allocating resources, managing resources among multiple projects/clients, execution of projects, updating progress to clients, monitoring project status and cost, providing feedback to team members, helping them achieve project deliverables under estimated time/cost by working on development stacks, providing testable modules to QA and getting quality assured, deploying project deliverables on servers, communicating with clients for feedback and follow-up for new business and other stuff required to deliver successfully projects to clients. Technology Stack: Laravel, Angular, Apache/Nginx, MySQL/MariaDB, PHP 5/7, WordPress, CakePHP, Yii, Zend (Laminas), Joomla, API integration Duration: 3 Years and 7 Months. 11-11-2013 to 30-June-2017. Organization : Cyber Infrastructure Pvt. Ltd, Indore. Job Profile: Team Lead, working on PHP, Javascript, JQuery, Zend (Laminas) MVC, CakePHP, Wordpress, Joomla, Dolphin, on Flash/Flex technology, Coding,Adobe Flex, ActionScript 3, AJAX, with modular approach and following standards, designing Database (MySQL), code review, developing OOPs based classes For Modules. Coordination with team members(developers, designers, QA) and work allotment. Interaction with client and project manager. Reporting progress, querying the client. Duration: 4 Years. 7-Oct-2009 to 30-10-2013. Organization : Intercom Online Pvt. Ltd, Indore (Anylinuxwork.com) Job Profile: Team Lead, working in LAMP environment, Coding (using PHP, Javascript, AJAX, Zend (Laminas) MVC, Savant, Joomla, OsCommerce, Dolphin) with modular approach and following standards, designing Database (MySQL), code review, developing OOPs based classes For Modules. Coordination with team members(developers, designers, QA) and work allotment. Interaction with client and project manager. Reporting progress, querying the client. Duration: 1 Year and 9 Months. From 3-Dec-2007 to 29-Aug-2009. EXPERIENCE AI/ML Engineer Self-Driven Projects & Freelance [Jan, 2023] – Present ● LLM Fine-Tuning & Embedding: Fine-tuned LLMs (OpenAI, Hugging Face) for domain-specific tasks; implemented custom embedding pipelines for semantic search and RAG systems using Sentence Transformers and ChromaDB. ● NLP Frameworks & Tokenization: Built NLP pipelines using PyTorch, Hugging Face, and LangChain; optimized tokenization for cost and performance in production LLM applications. ● Attention Mechanisms: Developed self-attention, masked attention, and multi-head attention modules from scratch in PyTorch; applied these in transformer-based models for text and multimodal tasks. ● Sentiment Analysis: Created sentiment classifiers using DSPy, OpenAI, and custom PyTorch models; deployed real-time sentiment analysis for customer feedback and social media monitoring. ● Text Summarization & Semantic Search: Built extractive and abstractive summarization tools; implemented semantic search engines with vector databases and advanced retrieval techniques (sentence-window, auto-merging). ● Supervised & Unsupervised Learning: Delivered classification, regression, clustering, and contrastive learning solutions; improved model accuracy and interpretability through iterative refinement. ● Model Distillation & Optimization: Applied knowledge distillation, quantization, and pruning to reduce model size and latency; automated prompt optimization with DSPy and custom scripts. ● Chatbots & Conversational AI: Developed multi-turn chatbots using OpenAI, Anthropic, and LangChain; integrated voice (LiveKit), web automation (Playwright), and RAG for advanced conversational agents. ● Vector Database & RAG: Created scalable, searchable vector databases from knowledge bases using LlamaIndex and ChromaDB; enabled multimodal and real-time search. ● Model Evaluation & Monitoring: Used RAG triad, BLEU/ROUGE, semantic similarity, and human evaluation; implemented automated testing and monitoring with TruLens, Arize Phoenix, and CI/CD pipelines. ● Python Development: Wrote production-grade Python code for AI/ML, data processing, web APIs, and automation; leveraged async programming for scalable, real-time systems. ● Understanding and Applying Text Embeddings with Vertex AI Developed text generation models using Vertex AI's Python SDK. Implemented semantic search capabilities for access control policies. Built scalable pipelines using Vertex AI and BigQuery for permission analysis. Experience with text embeddings for policy document analysis ● Cloud Platform Integration Hands-on experience with GCP infrastructure. Implemented authentication and authorization flows using Vertex AI. Worked with Google Cloud project management and location-based services Achievements LLM Fine-Tuning & Embedding: ● Reduced model size by 60-70% through quantization and pruning techniques ● Achieved 2-3x faster inference times with optimized attention mechanisms ● Cut API costs by 40-50% through token optimization and prompt engineering ● Improved retrieval accuracy by 35% using advanced RAG techniques NLP Frameworks & Tokenization: ● Built 15+ custom NLP pipelines using PyTorch, Hugging Face, and LangChain ● Optimized token usage by 30% through strategic chunking and preprocessing ● Reduced training time by 45% with efficient data loading and batching Attention Mechanisms: ● Implemented 3 types of attention (self, masked, multi-head) from scratch in PyTorch ● Achieved 25% better performance with custom multi-head attention implementations ● Reduced memory usage by 40% through optimized attention computations Sentiment Analysis: ● Built sentiment classifiers with 92% accuracy using DSPy and OpenAI ● Processed 10,000+ customer reviews with real-time sentiment analysis ● Reduced false positives by 30% through structured output validation Text Summarization & Semantic Search: ● Created RAG systems with 85% retrieval accuracy using advanced techniques ● Built semantic search engines processing 50,000+ documents ● Achieved 3x faster search times with optimized vector databases Model Optimization: ● Reduced model latency by 60% through quantization and distillation ● Cut deployment costs by 50% with optimized model architectures ● Improved throughput by 2.5x with efficient inference pipelines Chatbots & Conversational AI: ● Developed 5+ production chatbots with multi-turn conversation capabilities ● Achieved 200ms response times for voice agents using LiveKit ● Built autonomous web agents processing 100+ pages per session Vector Database & RAG: ● Created searchable vector databases with 1M+ embeddings ● Implemented advanced retrieval improving relevance by 40% ● Built multimodal RAG handling text, image, and video content Model Evaluation & Monitoring: ● Implemented automated evaluation reducing manual review time by 80% ● Built CI/CD pipelines with 95% test coverage for LLM applications ● Created monitoring dashboards tracking 20+ performance metrics Python Development: ● Wrote 50,000+ lines of production Python code for AI/ML applications ● Built 10+ scalable APIs using FastAPI and async programming ● Created 15+ reusable modules for common AI/ML tasks PROJECT HIGHLIGHTS 1. AI-Powered Business Intelligence Platform ● Built a RAG-based system for document, image, and video search using LlamaIndex, ChromaDB, and OpenAI. ● Integrated advanced evaluation and visualization tools for business insights. 2. Autonomous Multi-Agent Ecosystem ● Developed browser and voice agents with Playwright and LiveKit. ● Orchestrated multi-agent workflows using crewAI and LangGraph. 3. Content Creation Studio ● Automated text, image, and video generation using Hugging Face, OpenAI, and Gradio. ● Delivered structured outputs and brand-consistent content for clients. 4. Code Generation & Testing Platform ● Built code agents with smolagents and E2B; automated testing and evaluation with Arize Phoenix and GitHub Actions. 5. Conversational AI with Long-Term Memory ● Implemented long-context chatbots using Jamba, LangGraph, and Anthropic Claude. ● Enabled persistent memory and context-aware conversations. 6. IAM Policy Analysis System ● Built using Vertex AI for automated policy analysis ● Implemented anomaly detection for permission configurations ● Integrated with GCP security tools for continuous monitoring 7. Access Control Automation ● Developed ML models for permission alignment ● Created automated workflows for least-privilege access ● Implemented monitoring pipelines using Vertex AI Pipelines Achievements 1. AI-Powered Business Intelligence Platform ● Processed 100,000+ documents with multimodal RAG capabilities ● Achieved 90% accuracy in document retrieval and analysis ● Reduced search time from 30s to 2s through optimized vector databases ● Built evaluation framework with 15+ automated metrics 2. Autonomous Multi-Agent Ecosystem ● Developed 3 types of agents (browser, voice, memory) with 95% success rate ● Orchestrated 10+ agent workflows using crewAI and LangGraph ● Achieved 150ms latency for real-time voice interactions ● Processed 500+ web pages with structured data extraction 3. Content Creation Studio ● Generated 1,000+ pieces of content using multimodal AI models ● Achieved 85% brand consistency through structured output generation ● Reduced content creation time by 70% through automation ● Built 5+ specialized content generators for different use cases 4. Code Generation & Testing Platform ● Generated 500+ code snippets with 80% accuracy using AI agents ● Automated testing reducing manual review time by 90% ● Built security scanning catching 95% of potential vulnerabilities ● Created CI/CD pipeline with 100% automated deployment 5. Conversational AI with Long-Term Memory ● Implemented persistent memory handling 100+ conversation turns ● Achieved 3x context window using long-context models ● Built multi-modal conversations processing text, image, and voice ● Reduced conversation errors by 60% through memory management 6. Advanced RAG System ● Built sentence-window retrieval improving context relevance by 45% ● Implemented auto- merging reducing redundant information by 30% ● Created evaluation framework with RAG triad metrics achieving 88% overall score ● Processed 50+ document types with unified pipeline 7. Multimodal Search Engine ● Built cross-modal search handling 10+ file formats ● Achieved 92% accuracy in image-text matching ● Created contrastive learning improving similarity scores by 40% ● Built recommender system with 85% recommendation accuracy 8. Voice Agent Production System ● Achieved 200ms end-to-end latency for real-time voice processing ● Built scalable architecture handling 100+ concurrent users ● Implemented safety filters reducing inappropriate content by 95% ● Created monitoring system tracking 15+ performance metrics 9. Web Automation Platform ● Automated 50+ web workflows with 90% success rate ● Built structured data extraction processing 1,000+ web pages ● Implemented MCTS algorithms improving decision accuracy by 35% ● Created screenshot analysis with 95% accuracy in content detection 10. Model Optimization Pipeline ● Reduced model size by 70% through quantization and pruning ● Achieved 3x faster inference with optimized architectures ● Cut deployment costs by 60% through efficient resource usage ● Built automated optimization reducing manual tuning time by 80% SELECTED COURSES & TRAINING ● Attention in Transformers: Concepts and Code in PyTorch ● Automated Testing for L