fewer LLM input tokens after re-architecting a Gemini agent into an intent → execution pipeline
Amit Kumar Banerjee.
I engineer production AI agents multi-agent systems distributed backends data platforms
6+ years turning ambitious prototypes into fast, reliable systems — from Gemini agents at Google to 4G LTE tower software at Samsung R&D.
less tenant-onboarding time (80–90%) after moving a time-series platform to multi-tenant
misrouted calls prevented every year by an NLP transcript classifier (~2,500 hrs saved)
while leading EU Digital Markets Act compliance across pipelines & ML models
From prototype
to production.
I'm a Software Engineer at Google where I lead engineering on workplace AI agents and room-optimization backends.
I write production Gemini agents in Python and design multi-agent systems over the Agent-to-Agent (A2A) protocol. On the backend, I build distributed services in Java, Kotlin, gRPC and Spanner. Before that I built time-series analytics platforms, data pipelines and applied NLP systems at Google; C/C++ software for 4G LTE towers at Samsung R&D; and full-stack web apps at SkoolTree.
I care about the boring-but-hard parts: latency, reliability, privacy and the developer experience of the people who run the system after me.
AI Agents & LLM Systems
Gemini agents, multi-agent orchestration over A2A, tool-calling reliability, token & latency optimisation, autonomous triage.
Distributed Systems
Java · Kotlin · gRPC · Spanner · Protobuf. Secure RPC proxies & access policies.
Data & Applied ML
Time-series analytics, BigQuery pipelines, NLP classifiers, data-retention & compliance.
Full-Stack Roots
Node.js, PHP, React, MySQL, MongoDB, Redis, Android (Kotlin/Java) — I still enjoy shipping the whole thing end-to-end.
Where I've been
building.
Software Engineer III @ Google
AI Agents, Distributed Systems & Data Platforms · Hyderabad
Workplace AI & Distributed Systems
- Technical lead for a Python + Gemini AI agent that books and optimises meeting rooms across Google Chat and Calendar.
- Split the agent from one large prompt into a two-stage intent → execution pipeline — cut input tokens by 43% and improved tool-calling reliability and latency.
- Designed the cross-team multi-agent architecture over the A2A protocol so scheduling, room-booking and campus agents work across services.
- Built a distributed room-optimisation backend in Java, Kotlin, gRPC and Spanner that ranks and swaps rooms using live availability, priority and indoor walking paths.
- Shipped a secure RPC proxy and access policies for presence-aware booking — cleared privacy review with no extra OAuth prompts.
- Created an autonomous multi-agent bug-triage system that investigates production issues, catches regressions and drafts replies to cut on-call load.
Data Platforms, Analytics & Applied ML
- Led backend design and launch of a self-serve time-series analytics API (Java, SQL, Protobuf) — dashboard setup down to minutes with few-second load times.
- Migrated the platform to a multi-tenant architecture; automated onboarding cut setup time by 80–90%.
- Built an NLP pipeline classifying sales vs. support transcripts — preventing ~20,000 misrouted calls (~2,500 hours) a year.
- Led EU Digital Markets Act (DMA) compliance across data pipelines and ML models over two waves with zero outages; built automated data-retention pipelines.
- Built a cross-product user-journey analytics prototype that won funding for a full launch. Mentored an intern and junior engineers; conduct technical interviews.
Senior Software Engineer @ Samsung R&D Institute
4G LTE Network Software · Bengaluru (Remote)
- Wrote high-performance C/C++ powering 4G LTE telecom towers for domestic and international carriers.
- Refactored core codebases to reduce cyclomatic complexity and duplication — improving code-health metrics.
- Debugged product defects and release issues to keep carrier networks running without interruption.
- Built Python automation; used Jenkins, Perforce and Jira for image builds and releases.
M.Tech, Computer Science @ NIT Goa
Data Science · Machine Learning · Big Data
Software Engineer @ SkoolTree Technologies
Full-Stack School ERP · Kolkata
- Cut API response times by 50% by removing bottlenecks, trimming payloads and indexing MySQL queries.
- Owned end-to-end full-stack development across PHP, Node.js, JavaScript, MySQL, MongoDB and Redis.
- Designed custom REST APIs for the core school data-management system; integrated the CCAvenue payment gateway.
The stack I
reach for.
AI & ML
Backend & Systems
Data & Cloud
Web, Mobile & Tooling
Things I built
for the love of it.
ARTI — Digital Home Assistant
A voice-controlled robot assistant with a Raspberry Pi 3 "brain" and an Arduino Uno "brawn". Chats, learns, keeps to-dos, sends email, takes photos, streams live surveillance — and drives itself around. My B.Tech final-year project.
Trip Recommendation System
A hybrid recommender blending content-based and collaborative filtering (SVD, Pearson, cosine) on a dataset I surveyed from 70+ real users.
lEARN — E-learning App
Android app for self-paced learning with video courses, certificates, progress tracking and a virtual lab — Kotlin client, Node.js API.
Private Bank of India
Pure C++ console banking system backed by a hand-built double-hashed array of linked lists — designed to scale to a billion accounts. Hidden admin panel included.
Overdrive
Chat + music player in one Android app, built in a 4-day sprint for low-RAM phones that struggled to multitask.
Explore everything on GitHub
Prototypes, coursework and weekend hacks — from Android apps to data-science notebooks.
Education &
recognition.
Education
Awards & Certifications
Have an idea worth
building?
Whether it's AI agents, systems design, or just a good engineering conversation — my inbox is open.