AI & Backend Engineer

Aditya Dubey

Final-year B.Tech Computer Science student (AI & DS) doing the work of a backend/platform engineer. I like building the infrastructure other features sit on — data access layers, provisioning tools, and authorization — and I treat model output as untrusted input rather than a trusted collaborator.

Indian Institute of Information Technology Kottayam

B.Tech Computer Science, specialization in AI & DS

Aug 2023 – Apr 2027 · Kerala, India

View ExperienceEmail
Resume preview
Open in DriveDownload
GitHubLinkedInTwitterLeetCode

Experience

AI & Backend Engineer Intern · ZapData

May 2026 – Present

subsidiary of FinanceBuddha — On-site, Bengaluru

  • —Built and solely own Shield, a 2-company internal operations platform, replacing manual SQL requests to the data team with a self-serve text-to-SQL assistant over 160M+ Snowflake records — DeepSeek V4 Pro, schema-grounded prompting, WebSocket streaming, and Clerk-backed authenticated sessions.
  • —Built an LLM-driven query-to-segment layer on Zapiq that converts free-form user queries into executable business logic, constrained to a system-defined catalogue of attributes so generated definitions stay schema-valid and safe to run; owned the catalogue lifecycle (definition, versioning, resolution) backing it.
  • —Re-architected Zapiq's (B2B product) CSV export to execute inside Snowflake (COPY INTO → S3), removing API memory limits on exports of lakhs to crores of rows; automated the underlying IAM role, trust-policy and external-stage provisioning into Ferry, a Bun CLI that cut an error-prone multi-hour setup to <2 minutes, adopted by both engineering teams.
  • —Architected end-to-end RBAC for tenants and internal staff with a support-impersonation path enabling live customer debugging without credential sharing; shipped Zapiq's centralised notification system delivering in-app and email (Mailgun) events under high volume, a URL shortener powering cross-product deep-link routing between Shield and Zapiq, and WhatsApp/Messenger integrations.

Smart Contract Engineer — Real-World Asset Tokenisation (POC) · Zoop Money

Oct 2025 – Feb 2026

proof of concept with Arunabh Parihar, Co-founder — Remote

  • —Built Brickchain, a permissioned real-estate tokenisation POC on ERC-3643 (T-REX) — 12 Solidity contracts making non-compliant transfers impossible at the contract level by enforcing KYC and transfer eligibility on every mint, transfer and burn via OnchainID, IdentityRegistry and modular compliance rules.
  • —Modelled each flat as exactly one token unit via on-chain supply invariants and implemented the full escrow lifecycle (listing, settlement, cancellation, ownership transfer) with a dedicated attester role for legal holds; prototyped an eRupee-style settlement rail on Hyperledger Fabric and translated Indian/EU securities requirements into contract logic.

Projects

Ferry

GitHub

Infrastructure Bootstrap Engine

  • —Open-source engine that makes ordering-sensitive cloud setup safe and repeatable — 55 integrations across AWS (EC2, IAM, S3), GitHub and Snowflake, including cross-provider flows such as GitHub Actions OIDC → IAM role (removing long-lived CI secrets entirely) and Secrets Manager → GitHub secrets sync.
  • —Each integration is a folder of declarative steps the engine probes and plans before mutating anything: conflicts abort pre-mutation, rollback unwinds only what the run created in reverse dependency order (including on SIGINT/SIGTERM), and secrets go to 0600 reports, never stdout. 715 tests.
BunTypeScriptAWS SDKSnowflakeGitHub APIZod

Second Brain

LiveGitHub

AI-Powered Content Management System

  • —Knowledge-management platform that makes saved content retrievable instead of forgotten — semantic search, shareable collections, real-time collaboration and persistent multi-turn chat, with messages and citations atomically persisted and ownership checks embedded in data-access queries.
  • —Engineered a RAG pipeline on Cohere Embed v4 + Pinecone with per-user vector isolation and similarity-threshold gating that rejects low-confidence retrievals to prevent ungrounded answers; async multi-source ingestion (web, YouTube, X, Reddit) on a Redis queue with master–worker execution, deployed as a Dockerised Express API behind Nginx on EC2 with JWT auth and GitHub Actions CI/CD.
ReactNode.jsTypeScriptPostgreSQLRedisPineconeAWS

Achievements

Jigyasa

GitHub

UCO Bank × IIT Kharagpur PSB Hackathon

Finalist, Top 20 of 500+ Teams

  • —Real-time voice-clone detection on live calls at AUC 0.997 / 3.1% EER, using a streaming AASIST + XLS-R + AST + WavLM ensemble.
PythonFastAPIPyTorchReact

SatvaNetra

GitHub

CIDEcode Hackathon — CID Karnataka Cybersecurity Division

2nd Place, National Level (50+ competing teams)

  • —Led a 4-member team to build photo/video/audio deepfake detection at 95% accuracy, shipped inside the 48-hour window.

Skills

Languages

TypeScript · JavaScript · Python · Go · Solidity · C++ · SQL

Backend & AI

Node.js · Bun · Express.js · REST APIs · GraphQL · WebSockets · Prisma · FastAPI · Flask · RAG · text-to-SQL · embeddings · prompt engineering

Infrastructure & Data

Docker · Kubernetes · AWS (EC2, ECS, S3, Lambda, IAM, STS) · Nginx · GitHub Actions CI/CD · Prometheus · Grafana · PostgreSQL · MongoDB · Snowflake · Redis · Pinecone · Supabase

Blockchain & Frontend

ERC-3643 (T-REX) · OpenZeppelin · Hardhat · ethers.js · Hyperledger Fabric · React.js · Next.js · TailwindCSS