I design and ship production AI agents, multi-agent RAG pipelines, and LLM integrations for real enterprise workflows. My focus is constraint-first, structured-output AI that operates reliably inside production boundaries — not isolated demos.
AI Agent Engineer & AI Integration Engineer
Autonomous Article → Video AI Agent. Turn any article, URL, or story into a complete short-form video through an end-to-end automated pipeline.
Autonomous Article & Topic → Video AI Agent
Turns any raw topic, bedtime story, or article URL into an engaging, narrated short-form video through an automated multi-stage AI pipeline. The agent autonomously handles content analysis, scene scripting, neural voice narration, AI visual retrieval, and Remotion 4.0 React video composition — with zero manual editing.
Enter a topic, bedtime story, or article on the left. The autonomous agent pipeline will generate scripts, visual assets, neural audio, and render via Remotion.
The user provides a topic, story, or article URL. A content extractor cleans and structures context for Gemini with strict JSON schema storyboarding. For each screen, the agent generates neural narration (Google TTS), orchestrates visual assets, aligns word-level timestamps for animated captions, and renders the composite video via Remotion React 4.0.
Audio/Caption Synchronization: Naive speech-rate estimates caused drifting text. Solved by extracting exact audio chunk durations and normalizing words per timestamp.
Remotion Containerized Rendering: Rendered headless React video compositions via Chrome within resource-constrained Docker containers on Cloud Run.
Cold Starts & Quota Protection: Implemented per-visitor rate limiting, daily video caps, and persistent SSE status streaming to handle transient network interruptions.
Containerized with Docker, deployed to Google Cloud Run with autoscaling (0-N instances) and graceful error fallback if third-party visual APIs fail.
Production AI agents and internal engineering tools built over the past year. Every system is structured around deterministic contracts, strict constraints, and real developer workflows.
Built a 4-agent RAG pipeline (retriever → document analyst → creative director → storyboard engineer, plus a QA validator) that turns technical docs into narrated, animated explainer videos in the browser, featuring a zero-dependency frame-based animation and client-side MP4 export engine.
Built a zero-dependency Python backend agent that matches error signatures against curated reference playbooks and executes live GitHub code search across linked repos, returning match verdict, root cause, fix, and exact file references in a single query.
Authored an autonomous Python agent that crawls documentation portals with Playwright, validates each checklist item with an LLM, captures evidence screenshots, and compiles a shareable HTML/JSON audit report with confidence-scored verdicts.
Designed a custom VS Code agent that inspects request complexity, monitors remaining Copilot credit budgets, and advises developers on the cheapest safe model tier while maintaining full agentic capabilities.
Built a Google-AI-Overview-style search experience over an enterprise MkDocs portal, and engineered the client-side interaction-tracking context summarizer for a conversational banking UI that translates raw user events into compact state objects sent with every AI call.
Built full-stack UI and scalable APIs across three core Finacle banking modules — transaction tracking, expense approval workflows, and cross-border payment processing — serving 50+ enterprise banking clients worldwide.
Five foundational principles that separate predictable production engineering from unpredictable prototypes.
Every agent starts with strict boundaries of what it must NOT do. Define explicit operating scopes to eliminate hallucination and out-of-bounds behavior.
Ground every generation in trusted sources, curated playbooks, versioned documentation, and vector stores. Never rely on raw model pre-training.
Decompose monolithic prompts into specialized, cooperating agents (retriever, analyst, validator) with isolated context windows and distinct tasks.
Enforce strict Zod/JSON schemas and automated repair pipelines on every LLM call. Treat AI output as untrusted external input that must be validated.
Embed intelligence into existing developer and business workflows (CLI tools, IDE extensions, doc portals) rather than isolated chat windows.
6+ years of continuous growth across EdgeVerve Systems and Infosys.
AI Agent & Integration Engineering Focus
Bengaluru, India
Bengaluru, India
India
Every technical skill is backed by real project implementation — no self-assessed percentage bars.
Primary Focus
Enterprise Production
Deployment & Runtime
Interactive demonstrations of the core engineering patterns I use in production AI systems: Function Calling, Context Bounding, and Structured JSON Schema Enforcement. View public repository
This assistant uses Gemini Function Calling with 4 custom tool definitions. The model autonomously decides which UI tool to invoke based on conversation context. Candidate profile information is injected as bounded structured context.
Paste a Job Description (JD) to evaluate against my professional profile. The agent transmits both to Gemini requesting a strict JSON schema containing matched skills, experience alignment, gaps, talking points, and an objective fit score.
Paste any code snippet. The agent detects language, rejects non-code submissions via constraint boundaries, evaluates across 5 dimensions, and returns validated JSON rendered into actionable review feedback.
Selected public repositories and architecture implementations. Enterprise systems are maintained in internal organization repositories.
This portfolio codebase — demonstrating client-side Gemini function calling, Zod/JSON schema prompt contracts, rate-limiting edge proxies, and clean responsive architecture.
Autonomous article & topic-to-video generator containerized with Docker and running on GCP Cloud Run. Multi-stage script, neural voiceover, and Remotion React 4.0 composition.
I'm a Software Engineer with 6+ years of enterprise full-stack experience across EdgeVerve and Infosys, currently specializing in AI Agent and Integration Engineering.
I build multi-agent RAG systems, LLM-powered middleware, and autonomous engineering workflows that operate reliably within real production constraints. My focus isn't simply adding an LLM call to an application — I design the retrieval, orchestration, validation, and integration layers required to make AI systems predictable and valuable.
Multi-agent RAG pipelines (AI Canvas, Bedrock Claude, PGVector), autonomous QA/governance agents, and conversational UI context engineering.
6+ years shipping high-availability enterprise applications across React, TypeScript, Java/Spring Boot, and Node.js.
Strict schema validation, bounded context, graceful fallbacks, and reproducible deployment across AWS and GCP.
Interested in AI Agent & AI Integration Engineering roles. Reach out directly via email, LinkedIn, or schedule an interview discussion.