Senior Applied AI Engineer · Bengaluru
Seven years building production AI systems. My recent work spans agentic RAG over seismic data, local LLM inference on NVIDIA Jetson, and upstream open-source contributions.
Currently
Senior AI Engineer, Data Platforms
TGS (Steerwise) · Remote, India · Jul 2022 to Present
production-ai.systems
01 · Profile
I build agentic AI systems that retrieve evidence, use tools, extract structured data, and validate their outputs. At TGS, I apply that approach to multimodal search across petabyte-scale seismic data.
My foundation is in C++ and computer vision. More recently, I have worked on edge LLM serving with TensorRT Edge-LLM, continuous batching, chunked prefill, and measured latency and throughput on Jetson Orin. I work across data, model, evaluation, infrastructure, and runtime performance.
Build for evidence. Evaluation, observability, and benchmarking belong inside the architecture.
Engineer the full path. From raw, high-entropy data to a reliable production interface.
Optimize where it matters. Retrieval quality, latency, throughput, memory, and cost.
02 · Expertise
LangGraph workflows that connect retrieval, tool calls, structured extraction, and validation over complex enterprise data, backed by embeddings and IVF/FAISS vector search.
Faithfulness, retrieval precision, hallucination monitoring, drift analysis, and operational telemetry for latency, throughput, and cost.
Petabyte-scale ingestion and processing, workflow orchestration, cloud-native storage, data quality, Kubernetes, and event-driven systems.
Local Qwen3.5-4B serving on Jetson Orin, with continuous batching, chunked prefill, and request scheduling. Earlier edge vision work spans TensorRT, ONNX, DeepStream, and automotive SoCs.
Reproducible pipelines, CI/CD, model lifecycle tooling, automated retraining, and root-cause analysis across cloud and edge workloads.
03 · Selected systems
01 / Agentic AI
Combined ViT/MAE and LLMs with multi-step retrieval, tool calling, structured extraction, and validation over unstructured seismic metadata.
02 / Edge LLM
Submitted native two-sequence continuous batching and chunked prefill to NVIDIA TensorRT Edge-LLM. A qualified Jetson Orin test in the open pull request reports 22.54 to 33.97 output tokens/s.
03 / Local AI
Built a shared OpenAI-compatible model endpoint across Jetson, an RTX laptop, and an AWS GPU host, using TensorRT Edge-LLM and vLLM backends with documented deployment and benchmarks.
04 / Trustworthy AI
Built LLM evaluation with RAGAS and operational monitoring with OTLP/Grafana for faithfulness, retrieval quality, hallucinations, latency, throughput, and cost.
05 / Edge vision
Deployed driver monitoring, video analytics, and tracking across automotive SoCs and NVIDIA Jetson using TensorRT, ONNX, DeepStream, and INT8/FP16 optimization.
06 / Open source
Contributed a fix for COCO JSON conversion failures in HumanSignal's Label Studio Converter. The upstream pull request was merged in May 2024.
04 · Experience
TGS (Steerwise) · Remote, India
Via Affine Analytics, Jul 2022 to Jun 2024; direct with Steerwise, Jul 2024 to Present
Visteon Corporation · Bengaluru
Beyond Drops · Bengaluru
Built fingerprint biometrics and browser-based eKYC systems using deep learning, GCP, Docker, OpenPose, MediaPipe, and TensorFlow.js.
Integration Wizards · Bengaluru
Delivered real-time CCTV intelligence and custom multi-object tracking on NVIDIA Jetson using DeepStream, GStreamer, TensorRT, ONNX, CUDA, and optimized C++.
ApplicateAI · Bengaluru
Built conversational AI assistants with Rasa and spaCy, and developed demand-forecasting models using deep regression networks.
05 · Toolkit