DevHub
A social platform for developers to showcase projects and connect with each other.
- React
- Redux
- Node.js
- Express
- MongoDB
I build high‑scale, low‑latency systems. multi‑tenant AI platforms. distributed systems. products thousands rely on every day. AI that cuts costs, not corners.
I specialize in software architecture, distributed systems and generative AI, and I enjoy turning hard problems into simple, scalable systems. Right now I'm an SDE 2 and founding engineer at Meesho AI Services, building the core platform the vertical runs on.
Architected the core multi-tenant AI platform behind voice, chat, agent assist and campaign management. It now handles 2.5M+ interactions a day and made AI Services a new business line for Meesho.
0 → 30+enterprise clientsReal-time insights and guided actions for 2,000+ concurrent support agents, all in one interface, which also brought agent costs down.
−35%average call handling timeDAG-based orchestration on a config-driven CRM and data platform. 50+ product and business people launch agents themselves, with far less dependence on engineering, and new clients onboard much faster.
One platform for configuration, live A/B experiments and business-logic changes across 10+ services.
<1msoverhead at 1,000+ RPSRegion selection and cross-model fallbacks that cut mean latency by 200ms and let the platform recover from provider incidents 90% faster.
−75%LLM error rateReduced through a model experimentation platform and cross-model fallbacks, enabling faster evaluation and onboarding of new LLM, TTS and ASR models.
Automating tenant onboarding, testing and RCA feedback loops, halving go-live timelines and multiplying the pilots each person can deliver.
Customer support voice bot on a real-time ASR, LLM and TTS pipeline. Cost per call down 75%, handling time down 40%.
1M+calls a dayDAG-orchestrated chatbot with retrieval over support knowledge, handling 1M+ queries a day and routing each one to the lowest-cost capable model.
Migrated the end-to-end flow from a legacy PHP monolith to an event-driven Java microservice.
5M+requests a dayRebuilt the experimentation setup to run concurrent A/B tests on catalogue images, speeding up generative vision experiments for product discovery.
Operator dashboard for Eye of Satoshi, a Bitcoin Lightning watchtower, with real-time user, appointment and breach tracking. Also improved error handling in teos-cli.
100+operatorsQuery auto-fix and dry runs for a database-aware text-to-SQL tool, plus support for MySQL, PostgreSQL, BigQuery and Snowflake side by side.
85 → 94%SQL query accuracyWorked across 10+ digital insurance products and led six launches, with a front-end rework and client-side caching that made pages load faster.
Pick a focus area above to highlight what I use for it.
I'm a Software Development Engineer 2 at Meesho. I joined as an intern in 2024, helped launch India's first LLM-powered customer support voice bot at scale, and then became a founding engineer of Meesho AI Services.
I like problems where distributed systems and AI meet: routing across models and regions, keeping latency low on real-time voice, and building platforms that let other people ship agents without waiting on engineering.
A social platform for developers to showcase projects and connect with each other.
A sorting algorithm visualizer with custom inputs, speed and array-size controls.
A full-stack shopping website and clothing store management system.
A web app for finding nearby stores and facilities based on what you need.
A cross-platform mobile app for news and articles.
I'm always up for a conversation about systems, development, AI or anything in between.
ayushsaxena823@gmail.com