amit Kumar
Contact details — on requestProof of Work — on request
Experience
M
Principal AI Engineer
Matrimony.com · Oct 2025 – Present
- Architected a production-scale Agentic AI platform leveraging autonomous LLM agents and a 500+ rule validation engine to automate profile and photo moderation for millions of users, enforcing policies for contact information, competitor promotion, profanity, relevance, and unsafe content.
- Engineered a parallel LLM inference engine with asynchronous request execution, reducing end-to-end validation latency by 30% (to 1s) while increasing throughput for real-time moderation workloads.
- Provisioned highly available AWS infrastructure using Terraform, automating deployment of VPCs, ECS, EC2, S3, IAM, and networking resources, reducing infrastructure provisioning time from hours to minutes.
- Containerized AI microservices using Docker and implemented CI/CD pipelines with GitHub Actions, enabling zero-touch deployments and faster production releases while integrating PostgreSQL for profile metadata and validation logs.
- Implemented production observability using Grafana, monitoring API latency, throughput, GPU/CPU utilization, and infrastructure health with proactive alerting, maintaining 99.9% service availability.
- Designed domain-specific ML models including BLSTM classifiers for name-to-gender and name-to-religion prediction, and fine-tuned Google CANINE for robust character-level understanding of multilingual names, slang, and spelling variations.
- Reduced manual moderation effort through end-to-end AI automation, delivering operational savings exceeding INR 1.2 million per month while improving moderation consistency and user profile quality across production systems.
- Designed and deployed a production-grade Agentic RAG chatbot with hybrid retrieval (semantic + keyword) enabling context-aware responses and intent-driven action execution for complex user queries.
- Developed a modular agentic architecture supporting tool invocation, multi-step reasoning, persistent memory, and backend workflow automation, reducing resolution time for support queries significantly.
- Scaled with robust APIs, caching, and monitoring pipelines, driving measurable improvements in user engagement, response accuracy, and support automation in production.
Principal AI Engineer
Matrimony.com Limited · Oct 2025 – Present
- I currently work as a Principal AI Engineer at Matrimony.com, where I design and build large-scale AI solutions that improve profile and photo moderation using Agentic AI, Large Language Models (LLMs), multimodal AI, and cloud technologies.
- One of my key contributions has been leading the development of a real-time AI-powered validation platform that automates profile and photo moderation for millions of users.
- The platform uses autonomous AI agents and an extensive rule engine to detect issues such as contact information, competitor promotions, profanity, irrelevant content, and unsafe images.
- To support real-time performance, I optimized the inference pipeline with parallel and asynchronous processing, reducing response time by around 30%.
- I also built the cloud infrastructure on AWS using Terraform, containerized services with Docker, and implemented CI/CD pipelines through GitHub Actions for reliable and automated deployments.
- The platform is continuously monitored using Grafana, ensuring high availability and stable production performance.
- Beyond platform engineering, I have developed machine learning models for tasks such as name-to-gender and name-to-religion prediction, and fine-tuned Google's CANINE model to handle multilingual names, spelling variations, and slang effectively.
- These AI solutions have significantly reduced manual moderation effort while improving consistency and user experience, resulting in operational savings of over ₹12 lakh per month.
- Additionally, I designed and deployed an Agentic RAG-based chatbot that combines semantic and keyword search with intelligent agents to provide context-aware responses and perform actions based on user intent.
- The chatbot supports multi-step reasoning, tool integration, and scalable knowledge retrieval, enabling users to receive accurate answers and complete tasks seamlessly.
Tech Lead (AI/LLM ENGINEER)
HCL Tech · Oct 2023 – Oct 2025
- Developed an advanced document-based Q&A chatbot with follow-up question generation, source-linked answers, buffer/cache memory, and persistent chat history.
- Designed and orchestrated a data pipeline for Selenium-based approach, parsing, image-to-text extraction (Claude 3.0 Vision), and metadata enrichment, integrated into an Agentic RAG system by hybrid retrieval (BM25 + semantic search + reranking) with ChromaDB embeddings.
- Deployed the solution on AWS with FastAPI and Bedrock LLMs, achieving 96% accuracy and $2.4M projected cost savings.
- Fine-tuned LLaMA 3-70B using LoRA and QLoRA on a custom-curated dataset for domain-specific QA, enhancing response accuracy, contextual depth, and reducing hallucination in complex query scenarios.
- Developed a real-time Driver Distraction Detection system using a custom ResNet-34 CNN (99.16% accuracy) and MediaPipe for face landmark tracking, optimized with Intel OpenVINO for low-latency inference.
- Built a Streamlit-based UI with an integrated analytics dashboard and deployed the solution as Docker microservices for cross-platform, real-time driver monitoring and alerting.
H
Technical Lead
HCLTech – Engineering and R&D Services · Oct 2023 – Oct 2025
- At HCLTech, I developed an AI-powered document-based Q&A chatbot that enabled users to interact with enterprise documents through natural language and receive accurate, source-backed answers.
- The chatbot supported follow-up questions, maintained conversational context with buffer memory, and provided persistent chat history for a seamless user experience.
- I designed and implemented an end-to-end document processing pipeline that included web scraping with Selenium, document parsing, image-to-text extraction using Claude 3 Vision, and metadata enrichment.
- The processed data was integrated into an Agentic RAG architecture using hybrid retrieval techniques, combining BM25, semantic search, reranking, and ChromaDB embeddings to improve the relevance and accuracy of responses.
- The solution was deployed on AWS using FastAPI and Amazon Bedrock LLMs, achieving approximately 96% answer accuracy while delivering an estimated cost saving of $2.4 million.
- Additionally, I fine-tuned Llama 3 70B using LoRA and QLoRA on a custom domain-specific dataset, improving contextual understanding, enhancing response quality, and significantly reducing hallucinations for complex enterprise queries.
Skills 0 proven through work
Also works with
Semantic Caching ImplementationCost-Saving Initiative ManagementMachine Learning Model DeploymentFallback Mechanism DesignWork Planning and PrioritizationTeamworkProblem SolvingMentoring and Peer CoachingApplication DeploymentSoftware Development PracticesProject Architecture PlanningProject ConceptualizationLLM-as-a-Judge EvaluationTechnical TroubleshootingAnomaly DetectionBM25 Re-rankingCohere ReRanker UsageReciprocal Rank Fusion ImplementationCosine SimilarityChromaDB Vector Database ManagementVector Database UsageDocument ChunkingRetrieval-Augmented GenerationClaude DesignImage ProcessingAWS Textract UsageDocument ParsingUnstructured Data ProcessingOCR System DevelopmentSelenium Test AutomationData ExtractionETL Pipeline DevelopmentClassification ModelingDeep LearningPython MultithreadingLow-Latency System OptimizationApplication Performance OptimizationPerformance TestingConcurrency OptimizationConcurrent Workflow ImplementationAsynchronous ProgrammingRedis Cache ImplementationData ModelingFastAPI DevelopmentData ValidationScalable System DesignLarge Language Model ConceptsAI Agent Development
Proof of Work
Proof of Work
amit shares this with people who ask. You'll hear back either way.
Education
Master of Technology (M.Tech.), Artificial Intelligence
Indian Institute of Technology (IIT) Guwahati · 2021 — 2023
Bachelor of Technology (B.Tech.), Mechanical Engineering
M.J.P. Rohilkhand University, Bareilly · 2017 — 2021
Projects
Contact details
Contact details
amit shares this with people who ask. You'll hear back either way.
Ask about this candidate
AI responses may contain errors. Proof status and source information are provided by Proof of Skill.
A Proof CV — one profile, kept current, with the proof attached. Back to the portal