cv
Education
-
2021 – 2025 Ph.D. in Computer Science
Örebro University & WASP, Sweden - Thesis: Neurosymbolic decision-making with LLMs.
- Supervisors: Luc De Raedt, Pedro Zuidberg Dos Martires.
- Fully funded by WASP (Wallenberg AI, Autonomous Systems & Software Program), Sweden's largest research program.
-
2017 – 2019 M.Tech in Artificial Intelligence
Indian Institute of Science, Bangalore - Thesis: Active learning in sequence tagging.
- GPA: 8.10/10. Supervisor: Ambedkar Dukkipati.
-
2013 – 2017 B.Tech in Electrical Engineering
Birsa Institute of Technology, India - GPA: 8.03/10.
Research & Professional Experience
-
Jul 2025 – Present Research Scientist, Postdoctoral Researcher
FAIR, Meta Superintelligence Labs, London, UK - AIRA3 (Kaggle Gold Medal): Co-first author of an autonomous research agent that fine-tuned a 30B reasoning model end-to-end with no human in the loop, placing 8th of ~4,000 teams in NVIDIA's live Nemotron challenge — the first Kaggle gold medal won by an autonomous system. It also reached gold-level performance in another Kaggle competition translating 4,000-year-old Akkadian clay tablets into English, showing it can generalize across domains. [announcement]
- AIRA2 (NeurIPS 2026): Co-first author; identified the bottlenecks limiting AI research agents on MLE-bench and removed them by scaling agents across many GPUs. [pdf]
- AgentBox (FAIR Engineering Award): Built the execution infrastructure for running autonomous agents at scale across MSL, the backbone of multi-GPU AIRA experiments.
- Production Impact: Applied research agents to production GPU and accelerator kernel generation: 1.25–17× speedups on recommendation and LLM models, 27% lower kernel latency (internal benchmark), weeks-to-hours development.
- Data Collection, Annotation & Vendor Management: Led large-scale data collection and preference annotation with external vendors, used for post-training of models and LLM judges to improve the long-horizon ideation, planning, and judging abilities of AIRA2.
-
Aug 2024 – Feb 2025 Research Scientist Intern
FAIR (Meta AI), London, UK - AIRA1 (NeurIPS 2025 Spotlight): Developed research agents for MLE-bench, studying search policies, exploration, and generalization; open-sourced aira-dojo. [pdf] [code]
- SAM 3 (ICLR 2026): Owned the compositional concepts effort, leading large-scale data collection and building the evaluation for segmenting concepts with attributes and relations. [pdf] [code]
-
Jul 2022 – Dec 2022 -
Apr 2020 – Sep 2020 Data Scientist
Amazon Alexa-AI, Bangalore, India - Alexa Metrics: Led an investigation into why offline evaluation metrics failed to predict live Natural Language Understanding error rates in production, and designed a better offline metric that tracks live errors, raising offline–online correlation across 4 locales and 2 device profiles over the last 10 model releases.
-
Jun 2019 – Mar 2020 Research Associate
Statistics & Machine Learning Group, IISc Bangalore - Networked MARL (AAMAS 2020): Co-first author; developed networked reinforcement learning agents that learn an emergent language to coordinate traffic signal control, and analysed how the language is grounded. [pdf]
Honors & Awards
- First Gold Medal in a Live Kaggle Competition achieved by an Autonomous System (2026)
NVIDIA – Nemotron Model Reasoning Challenge (8th / ~4,000 teams, AIRA3) - Featured in Forbes, AIRA2: Breaking Bottlenecks in AI Research Agents [article] (2026)
- FAIR Engineering Award, AgentBox – infrastructure for scaling autonomous research agents (2025)
- Nominee, WASP Best Thesis Award (annual, Sweden-wide; winner announced Jan 2027) (2026)
- Spotlight Presentation, NeurIPS 2025 (AIRA1) (2025)
- Top Reviewer, NeurIPS 2022 and NeurIPS 2024 (top 8%)
- Best Project Award, B.Tech thesis (2017)
- GATE 2017, Graduate Aptitude Test in Engineering – top 0.1%
India's national exam for postgraduate engineering admissions
Selected Publications
- AIRA3: Autonomous AI Research Agents in Live Competition
M Josifoski*, K Hambardzumyan*, E Toledo*, R Hazra*, N Baldwin*, M Kuchnik* (*: equal contribution)
Live Kaggle Gold Medal, 2026 [announcement] - AIRA2: Overcoming Bottlenecks in AI Research Agents
K Hambardzumyan*, N Baldwin*, E Toledo*, R Hazra*, M Kuchnik*, M Josifoski* et al. (*: equal contribution)
NeurIPS 2026 [pdf] - AI Research Preference Models
TS Foster*, B Al Omari*, T Fu*, T Mann, C Domond et al.
Preprint 2026 [pdf] - Training AI Co-Scientists Using Rubric Rewards
S Goel, R Hazra, D Jayalath, T Willi, P Jain, WF Shen, I Leontiadis, F Barbieri, Y Bachrach, J Geiping, C Whitehouse
ICML 2026 [pdf] [dataset] - COvolve: Adversarial Co-Evolution of Large-Language-Model-Generated Policies and Environments via Two-Player Zero-Sum Game
A Sygkounas, R Hazra, A Persson, PZD Martires, A Loutfi
GECCO 2026 [pdf] - SAM 3: Segment Anything with Concepts
N Carion et al.
ICLR 2026 [pdf] [code] [website] - AI Research Agents for ML: Search, Exploration, and Generalization in MLE-bench
E Toledo, K Hambardzumyan, M Josifoski, R Hazra et al.
NeurIPS 2025 (Spotlight) [pdf] [code] - Combining Code Generating Large Language Models and Self-Play to Iteratively Refine Strategies in Games
Y Bachrach, E Toledo, K Hambardzumyan, D Magka et al., R Hazra
IJCAI 2025 [pdf] - LexiCon: Planning under Temporal Constraints in Natural Language
P Mantenoglou, R Hazra, PZD Martires, L De Raedt
NeurIPS 2025 (DB Track) [pdf] [code] - Have Large Language Models Learned to Reason? A Characterization via 3-SAT
R Hazra, G Venturato, PZD Martires, L De Raedt
COLM 2025 [pdf] [code] - REvolve: Reward Evolution with Large Language Models using Human Feedback
R Hazra*, A Sygkounas*, A Persson, A Loutfi, PZD Martires
ICLR 2025 [web] [pdf] [code] - SayCanPay: Heuristic Planning with LLMs using Learnable Domain Knowledge
R Hazra, PZD Martires, L De Raedt
AAAI 2024 [web] [pdf] [code] - EgoTV: Egocentric Task Verification from Natural Language Task Descriptions
R Hazra, B Chen, A Rai, N Kamra, R Desai
ICCV 2023 [web] [pdf] [code] - Deep Explainable Relational Reinforcement Learning: A Neuro-Symbolic Approach
R Hazra, L De Raedt
ECML PKDD 2023 [pdf] - Active2 Learning: Reducing Redundancies in Active Learning for Sequence Tagging
R Hazra, P Dutta, S Gupta, MA Qaathir, A Dukkipati
NAACL-HLT 2021 [pdf] [code] - Networked Multi-Agent Reinforcement Learning with Emergent Communication
S Gupta*, R Hazra*, A Dukkipati
AAMAS 2020 [pdf] - Workshop: LLM-Driven Adaptability or Pre-programmed Efficiency? A Comparative Study for Short Interactions
T Schreiter*, JV Ruppel*, R Hazra, A Rudenko, M Magnusson, AJ Lilienthal
HRI 2025 (Late-Breaking Report) [pdf] - Workshop: Bidirectional Intent Communication: A Role for Large Foundation Models
T Schreiter*, R Hazra*, J Ruppel, A Rudenko
LLMs in the RO-MAN Age @ IEEE RO-MAN 2024 [pdf] - Workshop: Intrinsically Motivated Compositional Language Emergence
R Hazra, S Dixit, S Sen
ViGIL @ NAACL 2021 [pdf] - Workshop: gComm: An Environment for Investigating Generalization in Grounded Language Acquisition
R Hazra*, S Dixit*
ViGIL @ NAACL 2021 [pdf] - * Equal contribution.
Mentorship
-
2024 – 2025 Master Thesis Supervision
Jens V Rüppel, TU Chemnitz - Co-supervisor: Tim Schreiter
Academic Service
- Session Chair: AAMAS 2022
- Program Committee: NeurIPS 2022–26 | ICML 2023–25 | ICLR 2024–25 | KR 2024 | EACL 2023 | TMLR