Machine Learning Engineer · 6+ years Bangalore, India

Machine learning engineer with 6+ years in product companies, shipping AI systems that hold up in production under real cost and latency constraints. Currently co-leading the platform-wide LLM copilot at Sumo Logic, owning long-term memory, agent evaluation infrastructure, and token efficiency. Partners across cross-geo engineering teams and directly with enterprise customers.

Cloud-native log management and observability platform · Nov 2024 – Present · Remote

• Agentic AI Copilot. Co-led design of the platform-wide LLM copilot serving 30K users across 5K enterprise customers. Owned context orchestration and the token-efficiency layer (summarization, trimming, prompt caching), cutting response latency 60% and inference cost ~30%.
• Long-Term Memory. Architected a shared memory system spanning org and user scopes, pairing human-in-the-loop saves with an observer agent that extracts and consolidates high-signal memories from live conversation. Hybrid retrieval (top-50 hot-memory cache plus agentic RAG over a file-like chunk interface) grounds 30% of user prompts and cut reference latency from 12 minutes to 4 minutes. Shipped across all agents.
• Agent Evaluation Platform. Took the eval suite from prototype to production grade: a capability-aware orchestrator routing agent output to an LLM judge scoring relevance, groundedness, and completeness, deliberately sampling low-confidence cases over random draws. Raised groundedness 0.70 → 0.85 and completeness 0.60 → 0.72 across all agents.
• Offline Eval and Regression CI. Built an offline eval pipeline from happy/sad-path cases mined from live judge metrics (no-data handling, exploratory queries, agent tone, multi-hypothesis reasoning), covering 70% of production prompt traffic and wired into CI/CD for regression tracking. A unified API/local eval interface on Phoenix cut the improvement cycle from weeks to days.
• Retrieval and Query Understanding. Owned a 2-second log-source detection pipeline using hybrid lexical and semantic retrieval (+50% downstream search throughput) and a sub-500 ms intent service that became the foundational primitive for conversational analytics (+70% query interpretation accuracy).

Solar digitization platform, Reliance Jio subsidiary · Feb 2023 – Nov 2024 · Remote

• End-to-End MLOps Platform. Co-led design of an automated MLOps system on AWS managing 5 production models across 45 client orgs, owning the full lifecycle (training, registry, versioning, serving, model selection, data and model drift detection) and collapsing ML cycle time 90%. Partnered directly with the CTO on ML direction.
• Production Computer Vision. Cut weekly construction-site reporting time 70% by deploying YOLOv8, U-Net, SAM, transformers, and GPT-based models over ~10 TB/week of drone imagery across thousands of solar sites.
• Distributed Integration Service. Built an event-driven OAuth 2.0 integration microservice (FastAPI, Airflow, AMQ/SQS, AWS S3, Lambda) powering 5K+ customer integrations and sustaining 1,000+ concurrent jobs via Microsoft Graph and AutoCAD.

Edge AI / MLOps platform for industrial computer vision · Jun 2021 – Feb 2023 · Hyderabad

• Edge Inference and Fleet Platform. Lifted on-device inference 25–30% via quantization (QAT, TFLite, TensorRT, OpenVINO) and structured pruning on resource-constrained hardware. Architected the web platform for real-time edge device operations (provisioning, telemetry, jobs, alerts, logs), cutting model management time 80%.

Central University of Karnataka — Integrated M.Tech, Computer Science (2015 – 2020) UG 84.9% · PG 95.8% · 3x Gold Medalist (UG, PG, University), awarded by ISRO Chief Dr. K. Kasturirangan.

Research Intern, MAVI — IIT Delhi (Aug 2019 – Aug 2020) Coordinated a 6-engineer team building an assistive edge device for the visually impaired: task allocation, systems integration, and professor liaison. Architected the inference pipeline for 40–70% faster on-device processing.