About
Applied AI Scientist with 3+ years of experience building and productionizing ML and GenAI systems across RAG, forecasting, smart-meter analytics, speech/audio, optimization, and model serving. Strong in Python, scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow, and HuggingFace Transformers, with production experience on FastAPI, Docker/Kubernetes, AWS, and PostgreSQL. Built hybrid retrieval, transformer-based TTS, and leakage-aware time-series/consumer-indexing pipelines.
Skills & Expertise (32)
Work Experience
Applied AI Scientist
Artha One
Jun 2025 - Present
Served as the MLOps bridge from model development to production inference, optimizing REST APIs and PostgreSQL schemas, adding caching/status polling, and reducing ML workflow query latency by 30%. Contributed to a LightGBM IEX Day-Ahead Market price forecasting pipeline at 15-minute resolution (96 blocks/day), engineering cyclic time, same-block lag, rolling-history, and regime features with chronological validation and production guardrails. Built a smart-meter data cleaning/evaluation pipeline over 7M+ rows, enforcing date/continuity rules, phase classification, and held-out phase-specific subsets for robust validation. Architected client-facing asynchronous ML workflows with Node.js/Express, PostgreSQL/Prisma job tracking, long-running Python inference, status polling, result/debug downloads, and validation.
Applied Scientist
Artha One
Apr 2024 - Jun 2025
Developed RF, RBF-SVM, and XGBoost consumer-to-DT matching experiments on aligned 24-hour voltage windows, engineering 30+ correlation, rank-correlation, error, lag, event, and phase features across five retained data splits. Productionized the final statistical inference path: 1-phase regression-style scoring by negative MSE and 3-phase Pearson-correlation voting across windows, aggregated by majority with phase/stability diagnostics. Formulated a Python MILP for non-smart consumer assignment using haversine proximity, load-capacity, and consumption-reconciliation constraints; validated assignments across Oct-Nov billing cycles. Developed production data/backend services used by thousands of users across social, e-commerce, and mobile products, strengthening API reliability and AWS deployment around ML-enabled workflows.
Software Development Engineer
Pinnacleworks Pvt Ltd
Sep 2023 - Apr 2024
Researched neural TTS architectures and trained/adapted a FastSpeech-family acoustic model on proprietary speech data with PyTorch/GPU tooling for an education voice bot. Built domain-focused speech data preprocessing by validating text-audio pairs and normalizing application numbers, dates, times, emails, course names, locations, and person names. Helped productionize speech inference with batching and FastAPI/TorchServe, S3 model versioning, Kubernetes worker scaling, and p50/p95/p99 latency/throughput dashboards in Grafana; reduced reliance on paid cloud TTS, saving ~INR 1 lakh/month.
Education
Bachelor of Technology - Netaji Subhas University of Technology
2019 - 2023 · Afghanistan
Certifications
No certifications added yet
Interested in this developer?
Profile Score Breakdown
Profile Overview
Availability Details
Visa Status
Citizen
Relocation
Open to Relocation
Skills (32)
Click a skill to find developers with the same skill
