13 systems, from GANs to agent platforms.
Open any project for the problem, the approach and the system diagram. Case studies include measured outcomes.
2026Agentic load-forecasting platformCase studyHARMAN International
Agentic AI platform for EV charging demand and utility grid load forecasting — time-series foundation models orchestrated by autonomous agents that handle data preparation, model selection, retraining, and anomaly triage, with forecasts exposed as callable tools for downstream planning agents.
Problem
Forecasts for EV charging demand and grid load have to stay accurate as data, seasons and usage shift — without a data-science team hand-tuning models every week.
Approach
- Time-series foundation models as the forecasting core
- Autonomous agents own data preparation, model selection, retraining and anomaly triage
- Forecasts exposed as callable tools so downstream planning agents can consume them directly
System
- Grid & charger data
- Prep agent
- Model-selection agent
- TS foundation model
- Forecast tools
Agentic AITime-series foundation modelsMulti-agentForecastingEV chargingSmart grid2024–2026Agentic AI platform for the software-defined vehicleCase studyKPIT Technologies
Enterprise Agentic AI platform with a LangGraph multi-agent framework for the automotive Software Defined Vehicle (SDV) lifecycle — orchestrating autonomous AI workflows across development, testing, validation, and deployment.
Problem
Automotive OEM clients needed AI across ADAS, in-cabin intelligence, connected vehicles and the full SDV lifecycle — including autonomous multi-agent workflows and on-device LLM inference on constrained SoCs.
Approach
- Multi-agent framework orchestrating autonomous workflows across development, testing, validation and deployment
- Enterprise RAG productivity tools on LangChain with GPT-4, LLaMA, Qwen and Claude
- LLMOps accelerator for orchestration, LoRA/PEFT fine-tuning, deployment and lifecycle management
- Led and mentored a 12-member engineering team
System
- Engineer request
- Orchestrator
- Dev · Test · Validate agents
- RAG knowledge
- Release
LangGraphLangChainGPT-4LLaMAClaudeQwenRAGPineconeFAISSLLMOps- Agentic AI interventions delivered
- 30+
- Agentic AI interventions delivered
- employee productivity gain from RAG tools
- 40%
- employee productivity gain from RAG tools
- engineers led and mentored
- 12
- engineers led and mentored
Voice AIReal-time voice RAG assistantCase study
Voice-enabled real-time conversational AI using RAG with LangChain, OpenAI Whisper (STT), TTS-1 and GPT-4. Low-latency streaming pipeline using WebRTC, AWS Lambda for parallel processing, and ElastiCache for caching.
Problem
Voice assistants feel broken the moment latency creeps past a second. The pipeline had to stream every stage — speech-to-text, retrieval, generation and speech — without waiting on the previous one.
Approach
- WebRTC streaming in and out of the browser
- AWS Lambda fan-out to process pipeline stages in parallel
- ElastiCache for hot retrieval and response caching
- RAG over a Pinecone index with LangChain
System
- Voice in
- Whisper STT
- RAG + GPT-4
- TTS-1
- Voice out
WebRTCWhisperGPT-4TTS-1LangChainPineconeAWS LambdaElastiCache- voice-to-voice response latency
- < 1 s
- voice-to-voice response latency
2020–2022Serverless CRM with conversational AICase studyLimitless Technology, UK
Highly secure serverless CRM for UK Fortune 500 clients. ML-powered conversational AI with NLP voice/text agents, a self-learning task dispatcher and real-time sentiment analysis, plus audio language conversion and closed-captioning for healthcare video streams.
Problem
Fortune 500 clients in healthcare and finance needed a CRM that automated routine work with AI — while staying secure and compliant in regulated industries.
Approach
- Fully serverless architecture integrating 20+ AWS services, deployed via CloudFormation and CodePipeline
- NLP voice/text agents with real-time sentiment analysis
- Self-learning dispatcher that routes, prioritizes and escalates tasks autonomously
- Live closed-captioning and audio language conversion for healthcare video
System
- Voice / text
- NLP agents
- Sentiment
- Self-learning dispatcher
- CRM
AWS LambdaConversational AINLPCloudFormationCodePipelinePythonNode.js- product revenue growth in Q4 2021
- 37%
- product revenue growth in Q4 2021
- AWS services in one serverless stack
- 20+
- AWS services in one serverless stack
2024–2026Enterprise LLMOps acceleratorKPIT Technologies
LLMOps platform for automotive AI — orchestration, fine-tuning (LoRA, PEFT), quantization (GPTQ, AWQ, GGUF) and deployment of LLMs including GPT, LLaMA, Qwen and Claude.
System
- Base model
- LoRA / PEFT
- Quantize
- Evaluate
- Deploy
LLMOpsLoRAPEFTQuantizationLangChainMulti-agent2024–2026In-car LLMs on Qualcomm SoCsKPIT Technologies
Fine-tuned and quantized LLMs (GPTQ, AWQ, GGUF) for on-device deployment on Qualcomm SoCs — ADAS copilots, in-cabin voice assistants and real-time contextual recommendations with low-latency inference on constrained hardware.
System
- Fine-tune
- Quantize
- Compile for SoC
- On-device inference
Qualcomm SoCGPTQAWQGGUFLLaMAADASIn-cabin AI2023–2024Real-time credit card fraud detectionYes Bank (via LTIMindtree)
Real-time fraud detection on transaction streams using Isolation Forest and XGBoost, with features on transaction velocity, geolocation and behaviour. End-to-end MLOps on the bank's on-premise Kubernetes + Kubeflow with A/B testing and drift monitoring.
System
- Transactions
- Features
- Isolation Forest + XGBoost
- Alert
XGBoostIsolation ForestKubernetesKubeflowMLflowPostgreSQLDocker2022–2024Customer churn predictionHDFC Bank (via LTIMindtree)
Churn prediction with ensemble ML (XGBoost, Random Forest, Logistic Regression) and time-series behavioural analysis across retail banking segments. MLOps on AWS with Snowflake and SageMaker, retraining automatically when performance degrades.
System
- Snowflake
- Behaviour features
- Ensemble model
- Retention signal
XGBoostRandom ForestAWS SageMakerSnowflakeMLflowpandas2022–2024Income classification for credit decisioningCapricorn Group, South Africa (via LTIMindtree)
Income classification model on Azure for automated customer segmentation and credit risk assessment, with Azure ML Studio and MLflow for experiment tracking, versioning, retraining and deployment.
System
- Customer data
- Classifier
- Credit decision
Azure ML StudioMLflowscikit-learnAzure FunctionsSQL2022–2024Multi-cloud MLOps acceleratorLTIMindtree MLOps CoE
MLOps accelerator spanning AWS, Azure, GCP and Databricks with standardized training, deployment and monitoring pipelines — plus an LLMOps platform for RAG and LangChain workflows. Featured in the most-viewed company blog of Q4 2023.
System
- Train
- Register
- Deploy
- Monitor
AWSAzureGCPDatabricksSnowflakeMLflowLLMOpsRead the write-up2019–2020IoT device management & data science dashboardVolansys Technologies
Full-stack dashboard for IoT device management and big-data analytics on AWS (S3, ECS, Glue, Athena) with Apache Airflow, PySpark, Dash-Plotly and React — real-time monitoring for HVAC and water-heater products.
System
- Devices
- Airflow + PySpark
- Athena
- Dashboard
AWSApache AirflowPySparkDash-PlotlyReactEdge AIHome-office automation & surveillance
Computer-vision models (TensorFlow, TFLite) deployed on NVIDIA Jetson Nano, with Azure REST APIs, CI/CD and an IoT cloud architecture on IoT Hub, CosmosDB and ML Studio.
System
- Camera
- Jetson CV model
- IoT Hub
- Automation
TensorFlowTFLiteNVIDIA JetsonAzure IoTCosmosDBApplied MLDCGAN face generation
Generator–discriminator Deep Convolutional GAN for generating realistic human faces, trained on 200k+ CelebA images with PyTorch.
System
- Noise
- Generator
- Discriminator
- Faces
PyTorchDCGANComputer vision
Have a system like this in mind?
Start a project