Work

Where the hours went

Three teams, three continents, one throughline — replacing manual operations with systems that run themselves.

Apr 2026 — now

MoveUp Media/AI Automation & Operations Engineer

Paris · remote

Automated an end-to-end content operations back-office over MongoDB — campaign orchestration, timed multi-channel publishing, AI image generation, and outcome settlement against a live sports feed, with zero manual intervention. Scaled the channel 200x, from 50 to 10k+ members.
Architected a 3-tier medallion data platform: raw JSON landed as Bronze in R2 object storage, normalised into Silver Iceberg columnar tables, projected into Gold relational tables in D1 — served to 5 downstream products behind a 14-endpoint REST API and edge cache, across 22 categories and 20+ sources.
Wrote the Go ingestion pipeline feeding Bronze: envelope-wrapped JSON on object storage, idempotent re-runs, full audit trail, 100+ records per run.
Built Python transformation pipelines projecting Bronze into 80+ column Silver tables on a 15-minute cadence — typed schema models, fuzzy-match deduplication, confidence scoring, 99%+ record reliability.
Shipped a 5-agent content engine on a serverless workflow runtime, auto-publishing structured articles to a fleet of sites and cutting manual production by 90%+.
Built 11 analytics dashboard views on React 19, including data-coverage, with configurable metric/dimension filters and CSV export.

Aug 2025 — Mar 2026

San Francisco · remote

Replaced manual lead workflows with an agentic discovery tool — extraction, normalisation, and structuring of listings from high-friction sites. Lead generation up 200%.
Built a dual-layer scraping engine: curl-cffi for low-level TLS/HTTP-2 fingerprinting, Playwright for human-like browser automation on JavaScript-heavy, high-security sites.
Engineered multi-tier escalation (HTTP → browser → proxy) with rotating residential proxies and autonomous CAPTCHA solving, routing intelligently to keep costs down.
Combined procedural parsing with AI-driven filtering for tight ICP alignment, and a fuzzy-match dedup + confidence scoring layer holding 99%+ record reliability with a full audit trail of domain success rates.

Oct 2024 — Jul 2025

Bangalore · remote

Engineered a sub-second conversational sales agent — an optimised streaming pipeline with native-tongue mimicry for automated outbound.
Designed and fine-tuned a multi-agent architecture that executes complex online tasks, including deep research, without continuous supervision — on custom tooling built from scratch.
Built an ICP research pipeline that runs interactive dialogues, converts natural language into structured JSON, and automates lead identification.
Built a lead enrichment system integrating web search, scraping, and AI relevancy analysis, plus a workflow generator for hyper-personalised email sequencing.

Education

B.Tech, Computer Science Engineering · Kolkata · 2023 — 2027

Stack

Languages: Python · TypeScript · Go · Rust · Solidity · Move · Java · C
AI / ML: Multi-agent systems · MCP · A2A · CoT / ToT / self-refine · deep learning · computer vision · fine-tuning · productionisation
Web3: Smart contracts · dApp architecture · tokenomics · ERC-20/721/1155 · gas optimisation · Foundry · Hardhat · wagmi
Data: Medallion architecture · Iceberg · R2 / D1 · Pandas · NumPy · PyTorch · TensorFlow · scikit-learn
Backend: Next.js · FastAPI · NestJS · Express · Django · SQL · MongoDB

Contact

If any of this overlaps with what you are building, say hello.