Archit Roy

Learner · Data Scientist · Python developer
Bengaluru, India

royarchit17@gmail.com | +91 9766687822 | LinkedIn | GitHub | Hugging Face

About

I build production data-science systems: pricing engines, propensity models, generative-AI workflows and the infrastructure that makes them reliable. I prefer understanding a problem deeply, explaining it clearly, and shipping something that survives contact with reality. Love learning new things, Growing and becoming better.

Experience

Selected work

• DS-NEMO: A production review-summarisation system built around latency, faithfullness, evaluation, observability and recovery. (Served 200K+ daily requests at sub-150ms p99 with no reported downtime over 10+ months.)
• DS-OSIRIS: A production propensity system—and an experiment that showed why prediction quality is not intervention quality. (The top decile converted at 3.16× baseline; experiments also revealed that targeting accuracy alone could not rescue a weak intervention.)
• DS-PIXEL: A rapid image-generation experiment that grew into a trust-preserving evaluation and scoring service. (Enhanced cohorts improved CTR in multiple city experiments while the evaluation layer filtered implausible edits.)
• DS-CIPHER: LLM-assisted support decisions grounded in calls, chats and trip context, with explicit audit and concurrency boundaries. (Auto-resolved about 51% of eligible tickets at 94% audited accuracy.)
• DS-WATCHDOG: A daily evaluation engine that separates unusual business movement from broken or incomplete data. (Converted an exploratory pricing workbook into a replay-tested, stateful daily evaluation service.)
• DS-PIGEON: A reusable reporting layer that turns successful analytical runs into readable, safe operational email. (Created a reusable bridge between pricing evaluation and the people expected to act on it.)
• PRICING: Base fares, asset controls, duration products, dynamic multipliers and change management treated as one ecosystem. (Reduced Goa basefare refresh cycles from weeks to under 30 minutes while keeping controlled override paths.)
• ALERTS: An analytical framework for deciding when a pricing movement deserves attention and what the receiver needs next. (Established a repeatable alert contract that reduces noise and makes silence interpretable.)

Technical skills

LANGUAGES

Python, SQL, C++

ML + STATS

XGBoost, Scikit-learn, SHAP, Pandas, NumPy, A/B Testing

GENERATIVE AI

Gemini, LangChain, RAG, LLM evaluation, DeepEval, Langfuse

SYSTEMS

Kubernetes, Docker, Kafka, Flask, Gunicorn, SQLAlchemy

DATA + CLOUD

BigQuery, MySQL, GCP, Vertex AI, ETL pipelines

OBSERVABILITY

OpenTelemetry, New Relic, Last9, Jenkins, ArgoCD, Signoz

Education

VIT Vellore — Bachelor of Technology in Information Technology, CGPA 9.09 / 10.0 (2021–2025)

Kunskapsskolan — CBSE, Grade 12: 94%, Grade 10: 91%

Certifications

• AWS Certified Cloud Practitioner (CLF-C02)
• Microsoft Power Platform App Maker Associate (PL-100)

Outside Work

• STORIES: Love a good story, be it movie, book or show.
• SPACE: Basicaly a way to get out of mundane life and think about the REAL bigger picture, the feeling of being in the grand scheme of things across trillions of years and multiverse and beyond.
• GROWTH: To Live is to Grow.
• SYSTEMS: Models with fallbacks, monitoring, ownership and real consequences. How to Build a self reliable perpetual system.

This page intentionally contains no interface, animation or visual trick. It is the record. Shout out to the real legends whose shoulders we stand on today.