Computer Science Student | Systems & ML Engineering

I build systems software, machine learning tooling, and data-intensive applications. My recent work centers on analytical query engines, ML inference runtimes, and distributed geospatial data pipelines.

Satellite oil-slick detection and vessel attribution platform (Smart India Hackathon 2026)

Problem & Approach: Automated detection of marine oil slicks from high-noise radar imagery, coupled with physical drift hindcasting to attribute spills to historical vessel transits.

Engineering Pipeline:

• SAR Segmentation: Ingests Sentinel-1 SAR imagery, applies adaptive Lee speckle filtering to suppress multiplicative radar noise, and segments candidate slicks via dynamic Otsu thresholding.
• Attribution Engine: Simulates leeway drift hindcasting using wind and ocean surface current vectors, correlating modeled slick trajectories with historical AIS vessel corridors to score candidate vessels.

Stack: Python, FastAPI, React, Leaflet, GeoTIFF, SAR Processing

Forward supply-chain forecasting and route optimization platform (SIH Problem Statement 26251)

Problem & Approach: Decision-support platform resolving forward supply-chain bottlenecks and dispatch routing across rugged terrain under dynamic weather and elevation constraints.

Engineering Architecture:

• Demand Forecasting: Generates weather-adjusted statistical demand forecasts, managing multi-tier delivery priority queues based on urgency, shelf-life, and storage capacity.
• Geospatial Routing: Leverages PostgreSQL/PostGIS to evaluate terrain elevation gradients and weather friction surfaces, producing cost-minimized, risk-scored dispatch itineraries.

Stack: Python, FastAPI, PostgreSQL / PostGIS, SQLAlchemy, React, TypeScript, Docker

Dedicated desktop numerical studio for time-series forecasting and quantitative backtesting

Problem & Approach: Native desktop quantitative studio in Scilab for model evaluation, comparing regression and smoothing models against naive baselines under transaction friction.

Engineering Implementation:

• Comparative Modeling: Evaluates Ordinary Least Squares (OLS) Linear Regression, Autoregressive (AR), and Holt exponential smoothing against random-walk baselines.
• Backtesting Engine: Executes expanding-window walk-forward validation with realistic transaction cost modeling, slippage parameters, and performance diagnostics.
• (Note: Dedicated Scilab GUI implementation, engineered separately from the earlier Python prototype).

Stack: Scilab (Numerics, Least Squares, GUI Engine, Plotting)

Measurement, Reporting, and Verification (MRV) platform for coastal blue carbon restoration

Problem & Approach: Hybrid on-chain/off-chain registry tracking ecological field measurements, ensuring verifiable record custody without offloading heavy telemetry on-chain.

Engineering Architecture:

• Off-Chain Pipeline: Mobile ecological field-data collection paired with a Node.js/Express and PostgreSQL backend for storing monitoring datasets and evidence documents.
• On-Chain Anchoring: The BlueCarbonRegistry.sol Solidity smart contract anchors cryptographic hashes of monitoring records and records whole-tonne credit issuance (1 credit = 1 tCO2e) with strict role separation (verifier vs. admin issuer).

Stack: Solidity, Hardhat, Node.js, Express, PostgreSQL, JavaScript

A curated selection of upstream contributions focused on fixing runtime bugs, standardizing CI test matrices, and handling core edge cases.

• apache/datafusion#26465 -- Passed CLI row counts into GCD function benchmarks in the DataFusion query engine.
• ionfwsrijan/PatchPilot#350 -- Implemented normalized severity ranking and deduplication in the security finding pipeline.
• ThreeMoonsLab/agents-shipgate#1007 -- Fixed parsing of PEP 508 extras-qualified pins during lockfile verification.
• batrapulkit/squidbrake#51 -- Established cross-platform GitHub Actions CI test matrices across Linux, Windows, macOS, and Python 3.13.
• knight22-21/CodePrism#72 -- Added strict configuration schema validation to reject unmapped settings and prevent silent fallbacks.
• corsairdev/corsair#1398 -- Integrated dynamic image asset generation plugin into the TypeScript framework core.

• ionfwsrijan/PatchPilot#354 -- Persists triaged finding status across page reloads (fixes #319). All backend, frontend, coverage, and security CI gates passing; awaiting maintainer review.
• safetensors/safetensors#873 -- Adds zero-length tensor slice handling in the Rust core to prevent panics on empty slices.
• Lightning-AI/pytorch-lightning#22006 -- Synchronizes sampler epoch state before iterator initialization for deterministic resumed training restarts.

View full contribution history on GitHub →

• XERJ Corpus Hub Contributor -- Included in the XERJ Corpus Hub reference catalogue for open-source work in mloda-ai/mloda.

• GitHub: Subodh-17
• Email: subodh171011@gmail.com
• Instagram: @subodh17_