Atharv
Electrical & Electronics Engineering undergraduate · Hybrid AI / Neuro-Symbolic Systems
Goa, India
@atharv_gaur
0 followers
B.E. in Electrical and Electronics Engineering at Birla Institute of Technology and Science, Pilani, K.K. Birla Goa Campus · Graduating 2028
About
Undergraduate researcher at BITS Pilani working on hybrid AI systems, multi-agent LLMs, and neuro-symbolic safety. Has authored accepted and under-review research publications and built reproducible, auditable ML pipelines and reasoning systems.
Experience
-
Undergraduate ResearcherBITS PilaniFeb 2026 – Present
Built the three-revision annotation pipeline behind a conflict-aware RAG benchmark and a multi-LLM judge committee; supervised open-weight models on evidence traces to improve refusal F1 across models and prompt settings.
Education
-
2028
B.E. in Electrical and Electronics EngineeringElectrical and Electronics Engineering (EEE) · Aug 2024 – May 2028
Skills
Tools / apps / platforms
Languages
Projects
-
Multi-Temporal JEPA for Satellite Image Time SeriesPyTorch, ViT, JEPA, Sentinel-2, Self-Supervised Learning
Built a causal temporal JEPA for Sentinel-2 crop series and improved representation quality with a variance-covariance regularizer.
-
Bid Intelligence over an Unstructured Document EstatePython, PyMuPDF, openpyxl, Deterministic Retrieval
Built an offline pipeline answering numerical business questions over 687 unstructured documents with byte-for-byte reproducible artifacts.
-
BDH Narrative Consistency EvaluatorPyTorch, Transformers, Mechanistic Interpretability
Built a train-to-inference pipeline for detecting contradictions across long-form novels with attention capture and heatmaps for interpretability.
-
LLM-Generated Control Barrier Functions with a QP FilterPython, CVXPY, CBFs, LLMs, A*
Built a neuro-symbolic safety stack where an LLM proposes executable control barrier functions and a convex QP filter enforces safety constraints.
-
Daedalus: Auditable Multi-Agent Scientific ReasoningPython, Multi-Agent LLMs, vLLM, pytest
Built a multi-agent LLM system that reads papers, extracts comparable claims, finds disagreements, and proposes falsifiable hypotheses with an auditable blackboard and tamper-evident ledger.
🏆 Achievements & awards
-
CUHK-X Multimodal Human Activity Challenge, UbiComp/ISWC 2026 – Top-15 Finalist · 2026
Selected for the Challenge Grand Finals and presenting a poster at UbiComp/ISWC 2026, Shanghai, Oct 2026.
-
LAM Research Challenge 2025 – National Finalist · 2025
Top 24 of approximately 7,000 registrations; robotics hardware build.
📚 Publications
-
The Floor Moves With the Probe: Trivial Baselines for Self-Supervised Learning on Satellite Imagery · 2026
-
From Evidence to Reasoning: Trace-Augmented Reasoning under Conflicting Evidence · 2026