Gautam Bhagat

Backend and AI engineer on Nudgebee, an open-source SRE copilot for Kubernetes/AWS/Azure/GCP — architecting the LLM agent pipeline, its data layer, and platform reliability.

• Architected multimodal support for the agent pipeline end-to-end: capability-gated vision-model routing, chat image attachments with persisted history, and a dedicated body-limit upload route.
• Engineered the conversation-context system powering long-running agent sessions — interval-based distillation, history summarization, and context propagation — plus a memory-fact extraction pipeline for durable recall.
• Optimized agent backend performance: parallelized per-label log-query validation to cut tail latency, and capped tool-call payload size to eliminate unbounded database growth.
• Built platform-wide backend infrastructure: Cost Analyser KPIs (cache hit rate, cache savings) and a generic multi-namespace cache-invalidation layer used across the product.

Backend and GenAI engineer on Integral, Lemma's ad-tech platform — architecting campaign management, forecasting APIs, and GenAI-driven ad tooling.

• Designed and built the Media Plan module end-to-end: versioning, an approval/rejection workflow, sharing, and role-based access control across sales and media-planning teams.
• Owned the Advertiser/Partner management APIs — dedupe constraints, multi-country geo-code support, and an ACL layer scoping access by partner account.
• Engineered a GenAI pipeline generating DCO ad creatives in 30–50s using Go channels for parallelism, and sanitized/cached the campaign forecasting engine's geo-based responses for speed.
• Revamped real-time campaign forecasting with LLM-driven insights, cutting prediction latency to under 4 seconds.

AI engineer building production computer-vision systems for physical-world access control and safety monitoring.

• Engineered a real-time face-recognition system for factory access control, authenticating 1,500+ employees.
• Built and deployed a PPE-violation detection pipeline with OpenCV, cutting weekly violations from ~100 to 15–20 within four weeks via real-time compliance alerts.

Backend engineer building the Django services powering a cab-based ad platform deployed across 150+ cabs.

• Built scalable Django backends handling 10K+ daily API calls for Greenvue Ads.
• Cut server response time 40% through async views and optimized endpoints.

Agentic AI runtime — tool-calling LLM agents with structured execution loops.

Core reasoning loop reused across personal agent experiments.

Async DAG workflow engine with dependency resolution, auto-resume, and dynamic task generation.

Handles long-running multi-step agent workflows without losing progress on failure.

MCP protocol chat backend with session management and Redis-backed conversation state.

Backbone for natural-language querying over UI dashboards and APIs.

Custom ReAct agent for HuggingFace with dynamic tool-calling and structured reasoning logs.

Reference implementation for debugging agent reasoning step by step.

Custom GAN-based dehazing algorithm integrated with a real-time drone feed over RTSP.

Improved visibility 40% and halved processing time for disaster-response operations.

AI exam platform with RAG-based question generation, real-time facial recognition, and tab-switch detection.

93% grading accuracy with automated integrity monitoring.

Lecture Notes in Networks and Systems · 2026

Smart Innovation, Systems and Technologies · 2024