I build production-grade machine learning and data systems that turn complex data into reliable products and measurable decisions. My work spans IBM Watson Studio and AutoAI, cloud data platforms, high-performance sports analytics, public-sector transformation, and full-stack AI product delivery.

7+ years in Data and AI · Ex-IBM Watson Studio & AutoAI · Peer-reviewed author · Production SaaS shipped · Upstream open-source contributions

Focus: Machine Learning Engineering · Data Engineering · Cloud Data Architecture · MLOps · Rapid Prototyping · Stakeholder Translation

Core languages:

• Email: palashcscs@outlook.com
• LinkedIn: in/pgds
• GitHub: PGupta-Git

• Saved ~15 hours/week and reduced data-retrieval latency by ~30% over 5+ years at IBM by scaling petabyte data pipelines, building forecasting models, and automating reporting workflows.
• Shipped a production SaaS with active paying subscribers: Drill Design App, built with Next.js, PostgreSQL (Neon), and Drizzle ORM (Case Study).
• Engineered a novel geometric metric from 25Hz optical tracking data and built a dual-signal player-similarity framework analyzing a full season of Premier League event data (BtLA Framework Case Study).
• Built non-linear models on high-frequency biometric telemetry and engineered predictive features alongside domain experts to replace legacy KPIs for player availability (Player Availability Case Study).
• Published a peer-reviewed randomized controlled trial applying ANCOVA and sensitivity analysis across six fitness outcomes (Paper · Data & Code).
• Led cross-functional delivery of decision-support dashboards and analytical products for performance monitoring, market sentiment, recruitment, and opponent scouting (Tactical & Recruitment Case Study).

Case studies are anonymised; organisation names and private code or data are intentionally omitted.

• Define the Decision First: Begin by identifying the business decision, user workflow, and measurable KPI. The target interaction dictates the system architecture.
• Design for Modularity & Speed: Build high-performance data pipelines and containerised cloud services alongside full-stack prototypes to accelerate iteration and feedback loops.
• Validate with Rigour: Establish robust baselines, define leakage-safe evaluation splits, and validate models using time-aware backtesting to ensure resilience against data drift.
• Communicate Uncertainty: Translate model outputs into explainable decisions, calibrating probabilities and handling edge cases gracefully rather than presenting black-box predictions.

Every contribution below links the reported problem, my contribution, and the verifiable upstream outcome.

• ageron/handson-mlp — community-endorsed Polars migration: Migrated all notebook exercises across 19 chapters and appendices to Polars, maintaining the polars_integration branch. The upstream author endorsed the community fork in Issue #23; PR #41 is open to add the upstream README pointer.

• posit-dev/positron — two fixes merged with credited authorship:

  • Console button alignment: Fixed native-button padding reported in Issue #14155. My PR #14381 was carried into merged upstream PR #14445.
  • Help search keyboard navigation: Added Enter and Shift+Enter match navigation for Issue #12921. My PR #14380 was carried into merged upstream PR #14444.

• can1357/oh-my-pi — five defect diagnoses translated into upstream fixes:

  • OpenCode MCP discovery: Diagnosed array-command coercion and ignored environment keys in Issue #3180, authored PR #3181, and received explicit credit when the equivalent fix merged in PR #3182.
  • Claude 4.6 provider routing: Reproduced wire-ID and token-limit failures in Issue #3067 and authored PR #3068. The maintainer's merged fix 47cc464 credits the diagnosis and includes a dedicated regression test.
  • Plan approval model retention: Identified the incorrect model-index comparison in Issue #3554; the exact proposed fix merged in PR #3556.
  • Linux desktop notifications: Diagnosed BEL fallback failures on VTE terminals in Issue #3685 and authored PR #3686; the production D-Bus solution merged in PR #3687.
  • Temporary model thinking overrides: Diagnosed the role-override failure in Issue #5290 and authored open PR #5291; the equivalent resolver and regression coverage merged in PR #5297.

• andrewRowlinson/mplsoccer — open visualisation features:

  • Speedometer and gauge charts: Ported and integrated the znstrider/speedo Speedometer class for Issue #16 in open PR #118.
  • Curved radar labels and resilient documentation builds: Added vector-glyph curved labels, multi-line layout, bottom-half auto-flipping, Wikipedia thumbnail standardisation, and request headers in open PR #120, fixing Issue #35.

• brilliantnz/flickernaut — merged stability fixes: Prevented crashes from invalid desktop files and removed duplicate preference entries in merged PR #9.

• oseymour/ScraperFC — regression identified and released: Reported missing FBref team and player identifiers in Issue #72; the maintainer verified the defect and released the requested fix in v4.1.0.

• Peer-reviewed research: A pragmatic, parallel-arm, randomised trial on the effects of two repeated-sprint training protocols on fitness outcomes in semi-professional male soccer players: preliminary report, published in Science and Medicine in Football (2026) · Data & Code Repository
• ORCID: https://orcid.org/0009-0000-0172-4009

Production focus: machine learning and data platforms, cloud-native delivery, reproducible analytics, and decision-support products.