Sahil Singh Architecting AI products. From 0 to 1 — and to scale.

Spine ships an AI research platform that produced #1-ranked deep-research outputs on Google DeepMind's DeepSearchQA benchmark. I owned the orchestration engine, model router, and large parts of the canvas runtime that made it possible.

• ▸Architected the backend orchestration engine for Spine Swarm — 1–10 specialized agents per task, 500+ concurrent, parallel non-sequential execution at production scale.
• ▸Built the intelligent model router — 400+ LLMs selected dynamically per execution step, optimising for quality, latency, and cost. Cumulatively processed 106k+ intelligence nodes.
• ▸Powered 6,400+ active canvas environments — collaborative agent execution, artifact persistence, branching, full inspectability of every agent run.
• ▸Designed and shipped APIs that produce client-ready outputs (decks, PDFs, Excel models with formulas) — not chat transcripts.
• ▸Built the React canvas frontend, scheduled-workflow system, and integrations across MCP, webhooks, and external APIs.
• ▸Owned multi-cloud DevOps (AWS / GCP / Azure) with zero-downtime blue–green deploys; took Spine 0→1 to 10k+ users · 151 paying · 2.5% conversion.

Founding-member SDE for the company behind Padhai.ai (UPSC, 175k+ users), EverTutor (K-12 + GRE), and a portfolio of AI tutoring products. Owned backend, infra, and reliability across the entire stack.

• ▸Owned the AI webserver behind Padhai.ai — comprehensive refactor, error handling overhaul, drastic stability gains during peak exam-season concurrent load.
• ▸Owned the backend & infra of EverTutor end-to-end — integrated Stripe to securely process international + domestic payments and unlock the entire revenue path.
• ▸Led migration from PM2-managed processes to a fully containerized ecosystem; designed blue–green deploys with Docker + PM2 supervisors for zero-downtime production updates.
• ▸Built a 10,000-concurrent WebSocket microservice for real-time AI interactions, alongside a CPU-intensive recordkeeping microservice — separated hot paths from heavy paths to keep latency low.
• ▸Architected Datadog APM + centralized logging across AI webservers and backends — proactive issue detection, faster outage response, full request traces.
• ▸Shipped a Redis-powered leaderboard with logarithmic-time access, enabling real-time rank surfacing without hammering the DB.
• ▸Built a markdown + LaTeX rich-text renderer in React Native — the runtime that displays every AI tutor response across products, including math/STEM-heavy content.
• ▸Architected, deployed, and managed containerized microservices across AWS / GCP / Azure — streamlined CI/CD, optimised cost, ensured high availability.
• ▸Led + mentored a team of interns; delegated quick-fix tracks to keep production velocity high.

• ▸Shipped React Native + Redux app features and Django APIs/websockets for AI chat + integrations — boosting app responsiveness and reliability.
• ▸Built a Redis cache leaderboard for in-memory logarithmic-time access; designed it to scale horizontally without warm-up cost.
• ▸Built a custom notification panel using Django + Firebase Cloud Messaging — bulk send, CSV upload, internal-tester segmentation, analytics, Celery-driven delivery.
• ▸End-to-end deeplink-driven sharing pipeline: dynamic image generation, S3 storage, Firebase deeplinks for previews — turned passive content into a growth surface.
• ▸Implemented compliant user-deletion that purges DB + cache; built hierarchical chat-graph viz with D3.js for AI conversation inspection.

• ▸Mission-critical POS flows in Next.js + Node, used heavily in production post-deploy across stores.
• ▸Shopify order sync via APIs/webhooks; Razorpay payments; SearchTap data + automated reindexing.
• ▸Discovered an indefinite OTP verification flaw in the GIVA mobile app and shipped a timeout-invalidation fix that hardened the auth flow.

• ▸Built an Incident Response GUI tool from scratch leveraging Volatility, Log2Timeline, RegRipper for forensic timelines and memory analysis.
• ▸Used MITRE ATT&CK CASCADE to map adversary tactics; built an ETL log parser for Windows pktmon/tracerpt → CSV pipelines.
• ▸Reverse-engineered WannaCry on Remnux, identified the kill-switch mechanism, built a malware lab to study live ransomware behaviour.

Maintained the org's website (frontend + backend), integrated Razorpay for payment processing in Node, shipped REST APIs and front-end improvements end-to-end.

Built a scoring + self-assessment web tool used by 400,000+ teachers across India. Pandas + NumPy scoring algorithm, Flask micro-framework, HTML/CSS frontend, oauth2/gspread for Sheets integration.