About
Problem solver by profession, problem creator by passion. I write code at SAP Labs, obsess over algorithms and math, and argue about time complexity at 2 AM for fun. When I'm not shipping software, I'm deadlifting 130 kg, squatting 120 kg, or chasing shuttlecocks on a badminton court — because my code isn't the only thing I push to production.
Experience
Leading the integration of generative AI-driven production scheduling and natural language interaction for complex supply chain tasks. Designing high-performance workflows to improve user decision-making in large-scale production planning. Collaborating on cross-functional AI initiatives to streamline enterprise resource management.
Engineered cross-module integration supporting approximately 20% of SAP's total revenue. Developed "Flexible Constraints," significantly increasing system efficiency and configurability for global enterprise clients. Orchestrated migration of legacy reporting infrastructures to modern systems, achieving a 3% reduction in processing time. Standardized deployment workflows via CI/CD pipelines on Azure, reducing average review time by two days per sprint.
Built Transformer-based NLP models for machine translation, improving contextual accuracy by 17%. Enhanced model performance through advanced linguistic feature engineering and transformer-based architectures.
Education
9.16 CGPA · National Institute of Technology, Silchar, Assam
Survived 4 years of algorithms, DBMS & operating systems — only to have AI threaten my job anyway.
Skills
Research
Published work on NLP and machine translation for low-resource Indian languages.
Sadhana (Springer), Vol. 47, 2022 · DOI
Read more about this research →
Projects
One-tap injured-animal reporting. Snap a photo → Gemini 2.5 Flash auto-scores species, injury and priority → PostGIS dedupe collapses same-spot sightings inside 100 m / 6 h → nearest vet sees it on a live map.
Upload a squat, pushup or deadlift video — get a skeleton overlay, per-joint angles, a state-machine rep counter, and coach feedback. MediaPipe PoseLandmarker runs client-side on WASM + WebGPU; the video never leaves your device. Optional Groq LLM for natural-language cues.
Semantic search + chat over your screenshots, photos and notes. CLIP image embeddings, FAISS index, Tesseract OCR for text extraction, and Gemini vision or Groq text for chat. HF OAuth sign-in with a persistent per-user bucket.
Satellite-based crop damage evidence for PMFBY claims. Fetches Sentinel-2 imagery before and after a weather event, computes NDVI change, correlates with Open-Meteo / IMD weather, and generates a timestamped PDF with severity + causal-confidence score for the farmer to attach to the claim.
Cross-platform Claude Code skills for repeatable engineering loops — PR review, repo review, release notes, bug repro, docs scaffold, dependency audit, weekly report, and 25+ more. Each skill is a self-contained slash command.
Persistent-Whisper voice input + wake-word command router for omp. Replaces a 5-second cold start per click with a long-lived worker — dictation lands in the editor in ~80 ms after warmup. Wake words like “computer, plan mode” dispatch slash commands directly.
macOS CLI + SwiftBar menu bar app that watches your INDmoney portfolio and watchlist via the official MCP server, with native notifications when things move.