Headline

AI Engineer building production LLM agents & RAG systems

About

AI Engineer with 3 years of experience shipping production-grade GenAI, LLM, and NLP systems — not just prototypes. I build things that survive contact with real users and real scale. What I've delivered: Cut LLM hallucination rates from 20–30% down to 2–5% using multi-agent RAG workflows (LangGraph, FAISS, Qdrant) — currently serving 500–2,000+ students in production Built a Voice Journal AI agent processing audio for 5,000+ users, with automated scoring across fluency, grammar, and confidence Achieved 90%+ accuracy on multilingual transcription using Whisper, cutting manual audit time significantly Saved ₹3L+/month through workflow automation; reduced reporting time by 40% via self-serve dashboards Core stack: RAG pipelines (FAISS, Qdrant, Pinecone) · Multi-agent systems (LangGraph, CrewAI, AutoGen) · LLM fine-tuning & prompt engineering · FastAPI · AWS, Docker · Speech AI (Whisper, diarization) I care about the gap between "works in a demo" and "works in production" — evaluation pipelines, hallucination grounding, and cost/latency tradeoffs are where I spend most of my time. Open to AI Engineer, GenAI Engineer, Applied AI, NLP Engineer, and ML Engineer roles. 📫 prateek02kumar@gmail.com 🔗 Portfolio: prateekportfolio-r2gd.vercel.app

Location

Bangalore Urban

AI stack

claude code

claude

codex

langchain

mcp

ollama

Work experience

AI/ML Engineer at QBS Learning, Sep 2024 - Oct 2026 · 2 yrs 2 mos, Bengaluru, Built and deployed curriculum-aware K-12 chatbot agents (Key Stages 1–5) using LangChain and LangGraph, serving 500–2,000+ students with personalized learning paths Cut LLM hallucination rate from 20–30% to 2–5% by designing multi-agent workflows with FAISS/Qdrant vector grounding and structured context memory Built a Voice Journal AI agent for 5,000+ students, automating audio cleaning, noise profiling, transcription, and multi-parameter scoring (fluency, grammar, confidence, topic relevance) Built a Market Research AI agent using web scraping and structured JSON pipelines to extract competitor pricing and marketing intelligence, replacing manual analyst workflows Fine-tuned small language models (SLMs) on STEM datasets, achieving 85%+ precision and cutting manual content creation time by 70% Defined response quality metrics, task success KPIs, and LangSmith evaluation pipelines for continuous production monitoring Prototyped 3 GenAI solutions (intelligent search, summarization, curriculum transformation) using GPT and LangChain — 2 shipped to production

Data Scientist at ffreedom app, Jan 2024 - Sep 2024 · 9 mos, Bengaluru, Built multilingual transcription pipeline with Whisper + LLMs, reaching 90%+ accuracy Designed speaker segmentation workflows to support compliance review Implemented a RAG system for internal knowledge retrieval Built AI-based QA pipelines that reduced manual audit workload Built Superset dashboards that cut reporting time by 40%

Data Analyst at Reloy, Apr 2023 - Oct 2023 · 7 mos, Bengaluru, Karnataka, India, Reloy, a RERA-compliant service provider is world's largest homeowner loyalty platform. Through its digital amenities, it helps Real Estate companies. ◆ Automated Prop-Tech client loyalty program using Beautiful Soup and Selenium, saving 3 lakhs PM in human labeling costs. ◆ Streamlined workflows with Zapier, improving operational efficiency and reducing manual labor. ◆ Tech Stacks: Python, SQL, AWS, Zapier , Machine Learning, Beautiful Soup, Selenium Web driver

AI/ML Core Member at GDSC CDTU, Jun 2022 - Jun 2023 · 1 yr 1 mo, Bengaluru, ◆ Actively contribute to the AI/ML community within the organization. ◆ Collaborate with a team of enthusiasts to explore and promote advancements in Artificial Intelligence and Machine Learning.

Data Science Intern at Mvg Innovations, Feb 2022 - May 2022 · 4 mos, Hyderabad, Telangana, India, ◆ Analyzed voter patterns and conducted EDA for regional political parties during election campaigns, uncovering key trends to guide strategy. ◆ Developed predictive models using machine learning and deep learning, achieving 92% accuracy in forecasting election outcomes compared to traditional methods. ◆ Built dynamic dashboards in Power BI and Tableau, delivering critical insights to facilitate informed decision-making for political campaigns. ◆ Tech Stacks: Python, Scikit-learn, Machine learning, Deep learning, Power BI & Tableau

Education

Master of Science - MS, Data Science, Christ University, Bangalore

Bachelor of Science - BS, Data Science, Christ University, Bangalore