Computational researcher working on reproducible AI, programmatic LLM evaluation, and benchmark integrity.
ssh ssh.badha.is-a.dev -p 2222
I conduct research at the intersection of AI-first development and verifiable computational methodologies. The work encompasses building applied artificial intelligence systems, including large language models and computer vision for tasks ranging from automated program repair to remote sensing. This development is anchored by rigorous empirical research. Programmatic evaluation and reproducible pipelines ensure all systems remain transparent and scientifically sound.
Superellipse-based localization method for satellite ship detection. Benchmarked on ShipRSImageNet (10,043 samples); mean IoU 0.5609 vs 0.4929 for OuterHBB.
In Smart Computing Techniques in Industrial IoT, Studies in Computational Intelligence, vol. 1172, Springer.
International Journal of Research in Humanities, Arts and Science.
Investigated grammar and phonology induction in low-resource languages using in-context learning with Qwen2.5-14B-Instruct. Designed inference strategies focused on calibrated prompting and minimal architectural complexity, achieving chrF 0.2998 on the IOL-AI 2026 benchmark. Demonstrated that careful prompt engineering with a single forward pass can match or exceed complex multi-stage approaches.
Multi-sensor satellite pipeline for ship detection, heading estimation, and dimensional measurement across Sentinel-1 SAR, Sentinel-2 optical, and PlanetScope sensors. Hybrid heading estimation combining border wake analysis, NIR wake detection, and ML-based prediction. Trained sensor-specific models with confidence scoring, deployed operationally at SkyServe for maritime surveillance.
Dual-brain reaction prediction engine fusing a 1.2B-parameter ReactionT5 Transformer (ONNX-optimized in Rust) with textbook functional group matching from a curated knowledge base. Delivers explainable predictions with confidence tiers — from pure ML inference to rule-verified results. Deployed as a high-performance axum REST API with PubChem enrichment.
Collaborating within a core four-member team to design architecture and curate datasets for a truly open video world model, establishing transparent open-science methodologies from day one. Co-engineering reproducible experimentation and analysis pipelines across diverse video datasets. Contributing to a global open-research ecosystem building scalable, publicly accessible AI and open-source codebases.
Engineered computer vision pipelines for remote sensing, developing high-precision convoy and ship detection models that secured a Top-3 national finish at the AIGC India Finals. Developed a novel Shape-Aware OBB to HBB conversion algorithm for satellite ship localization, improving mean IoU from 0.4929 to 0.5609 on the ShipRSImageNet benchmark. Architecting an automated road health assessment algorithm, optimizing geospatial vectorization workflows (GeoPandas) on high-performance clusters for scalable data processing.
Engineered a public-facing multilingual geospatial chatbot integrating programmatic LLM APIs to democratize access to complex data for non-technical users. Designed an LLM router for querying disparate geospatial data sources, ensuring transparent and verifiable data retrieval metrics. Developed empirical systems for road identification and solar-energy analysis, establishing baseline accuracy thresholds for public-sector deployment.
Architected an LLM-driven automated program repair pipeline, rigorously evaluating model performance via API against established software engineering benchmarks including SWE-bench Lite. Investigated benchmark contamination and data leakage in LLM training corpora, directly addressing trustworthiness and evaluation reliability in AI research. Expanded the Defects database, curating high-quality, reproducible evaluation datasets for empirical software engineering studies.