Vansh BTrust & Safety Data Analyst

Trust & Safety, Data AnalystBase version

What datahides, Isurface.

I validate the data decisions are made on: 6B+ rows reconciled from multi-source feeds, statistics analyzed about production ML systems, and failure modes surfaced before anyone had to report them.

6B+

rows validated across feeds

25+

ML systems evaluated end to end

500GB+

raw data parsed into analytics

About me

I am Vansh Bordia, a data analyst who hunts for what data hides. At RiftWatch I traced dirty ingested data to its source, built the QA checks and validation so the 6B+ row analytical layer could be trusted, and analyzed statistics about 25+ production ML systems, which is how I found hidden failure modes where the deployment pipeline silently passed empty fixtures through to hosted models. Nobody had reported them. The monitoring did.

I believe trust is built before the incident, not after. So I build the automated flows that catch problems early, I present findings with the artifact so stakeholders can verify rather than take my word, and outside of work I built Combat Corona, a COVID-19 awareness game awarded by the Government of India. Protecting users is not a slogan for me, it is a track record.

Why me

I do the unglamorous work that makes trust possible: tracing dirty data to its source, validating schemas across multi-source feeds, analyzing statistics about production model performance, and surfacing anomalies proactively before stakeholders have to report them.

Studying Data Science and AI at IIT Madras while doing the exact work: parsing 500GB+ of raw feed data into validated analytics, analyzing statistics about the performance of 25+ production ML systems, and surfacing hidden failure modes proactively rather than waiting for someone to report them. I built automated validation flows so problems get caught by systems, not by people, and I present findings so stakeholders can act fast.

Validated and reconciled 6B+ analytical rows from multi-source feeds with schema and QA checks

Analyzed statistics about 25+ production ML systems and surfaced hidden failure modes silently passing bad data

Built automated validation flows, catching problems by system design rather than human vigilance

Defined scope, goals and deliverables leading projects across 5 teams and 10+ SMB client engagements

Combat Corona, a COVID-19 awareness game, awarded by the Government of India

Role stack

SQL / Python / Statistics / Trend Identification / Summary Statistics / Data Validation / Schema Checks / Multi-Source Reconciliation / Anomaly Surfacing / ML Systems / Model Evaluation / Precision & Recall / Classification / Automated Flows / Power BI / Google Analytics / Stakeholder Presentation

Relevant work

RiftWatch

Esports intelligence platform, acquired by a pro org

  • 750+ pro matches processed into 6B+ analytical rows in ClickHouse
  • 20 orgs at peak including Tier-1 across Americas and EMEA
  • VALORANT-specific Gemma models plus RAG over live patch data
  • Agentic harness routing tools, models and queries; sub-500ms P90
PythonFastAPIClickHousePostgreSQLRedisKubernetesReact
Visit project

VLR Dev API

Type-safe Python library for Valorant data, published on PyPI

  • Type-safe Python library for Valorant data, published on PyPI
  • Retry logic, rate limiting and schema validation built in
  • Earned official Riot data API access in 2 weeks, not 9 months
PythonPyPIREST
Visit project

Combat Corona

COVID awareness game, awarded by Govt. of India

  • COVID awareness game built during the pandemic
  • Recognized and awarded by the Government of India
Game DevUnreal Engine
Visit project

Contact

Fighting abuse needs people who trust their data. Talk to me.

GitHubLinkedInBehanceDribbbleKaggleInstagramTwitterFacebookYouTubeTwitchSpotifyReddit