Anubhav Tandon
Open to work|
I build production LLM applications end to end: AI agents with tool calling, RAG, speech-to-text, LLM evals and multi-tenant backends on AWS. Founder of Plate8, an AI restaurant operations platform in pilot at 3 restaurants. Previously shipped GPU cloud backends and DevOps automation at Skyportal.ai & Oblivus Cloud.
Tech Stack
Professional Journey
From GPU cloud infrastructure to founding an AI startup. Click a role for the full details.
AI back-of-house operating system for Indian restaurants: voice-driven inventory, stock tracking and demand forecasting.
India
- Founded Plate8 and built the product end to end; onboarded 3 restaurants, currently under pilot, using it for daily inventory counts and stock tracking
- Lowered food wastage by 30% at pilot restaurants through efficient demand and stock forecasts
- Built a multilingual (Hindi, Punjabi, English) voice inventory pipeline: speech-to-text, LLM extraction with Claude Haiku, item matching and confidence scoring, targeting 90%+ line accuracy and under 3 s p95 latency at ~INR 145 per outlet per month
- Built Walter, an AI agent on the Anthropic Messages API with tool calling; every model and tool call is traced with cost and latency, and quality is measured by offline eval suites
- Designed a multi-tenant FastAPI + PostgreSQL backend with row-level security, an append-only stock ledger and event outbox, plus a React Native (Expo) app and Next.js dashboard
Skyportal.ai & Oblivus Cloud
Software Engineer (Contract)
Open Science Labs
Software Development Intern
Featured Projects
Production AI products I've built end to end, plus selected earlier work.
Plate8
Voice-first inventory and stock platform live in pilot at 3 restaurants. Staff speak counts in Hindi, Punjabi or English; a speech-to-text + Claude Haiku pipeline extracts, matches and confidence-scores line items, and Walter, a tool-calling agent on the Anthropic Messages API, answers operational questions with every call traced for cost and latency.
- 90%+ line-accuracy target, <3 s p95 at ~INR 145/outlet/month
- Offline evals: golden voice set + 200-phrase item-matching suite
- Multi-tenant FastAPI + Postgres with RLS, append-only stock ledger, event outbox
Qarz.ai
StealthAI document-processing pipeline that classifies borrower documents, extracts fields with confidence scores and runs 20 validation checks, 7 cross-document rules and 5 tamper detectors, gated in CI by accuracy thresholds. Full-stack platform with lender-rule checklists, consent-based borrower links, malware-scanned and Aadhaar-masked uploads, officer review and WhatsApp reminders.
- CI-gated extraction accuracy (field accuracy ≥ 0.90)
- 20 validation checks · 7 cross-doc rules · 5 tamper detectors
Open Source Contributions
Merged and in-review pull requests to production open-source projects.
Flagsmith
Flagsmith · Django feature-flag platform
Fixed project-admin custom field permissions, added soft delete for ClickHouse identities, and added multivariate environment diff detection.
Education & Achievements
Indian Institute of Technology (IIT) Jodhpur
Oct 2022 – May 2026B.Tech in Computer Science and Engineering
JEE Advanced 2022
Top 1% of ~1.2 million candidates
Codeforces Specialist
Global ranks 1523 (Div. 2), 1655 and 1673 (Div. 3); 500+ DSA problems solved
Get in Touch
Have a project in mind or just want to chat? I'd love to hear from you.
Open for freelance work, full-time roles, and interesting collaborations. Whether it's an idea, a question, or just a hello — feel free to reach out.