icon Tr$^2$AIL: Trust and Transparency in AI through Logic

The Tr$^2$AIL research group is led by Prof. Somak Aditya. The long-term goal of the lab is to achieve transparent Human-aware AI systems through Logic and Reasoning. We want the AI systems to be able to reason-in-the-wild and do so transparently for the end-users with whom it is expected to interact. Technically, we will work on the fundamental challenges faced by current neural and probabilistic logical systems in achieving such properties. I am fortunate to be working with a group of passionate student reserachers and my wonderful collaborators. Our group acknowledges the kind and generous support by Microsoft, Rakuten (through sponsored internship), Toloka AI, SERB DST, AI4CPS IIT Kharagpur (see Grants for details).
Group Members External Collaborators Publications Grants Codes/Datasets FAQs
1 / 6

2 / 6

3 / 6

4 / 6

5 / 6

6 / 6

❮ ❯


Grant Support

We are grateful to be supported by the following generous organizations: ANRF (erstwhile SERB), Microsoft Agentic Research Initiative, Microsoft AFMR, Rakuten, Toloka, IIT KGP AI4CPS and IIT KGP Faculty Startup Grant:
  1. ANRF ARG (2025-29)
    ARG (2025-29) ~ INR 96Lakhs | Topic: PrahelikaAI: Democratizing Education through Neurosymbolic Puzzle solvers in an Indian Context
  2. Microsoft AARI Azure+Cash Grant (2025-26)
    AARI Azure Grant (2024-26) ~ USD 206,000, USD 20000 (Cash) | Topic: Agentic Verifiers: Provably Safe Test-time scaling for Reasoning Models
    Co-Pi: Sourangshu Bhattacharya (IIT KGP), Uma Satya Ranjan (IIT Jammu), Amit Sharma, Microsoft Research India
  3. Microsoft AFMR Azure Grant (2025-26)
    AFMR Azure Grant (2024-26) ~ USD 8000 | Topic: Education Video Copilot: AI Teaching Assistant for Remote & MOOC Courses
    Co-Pi: Tanuja Ganu, Microsoft Research India
  4. AI4CPS IHUB Grant (2024-26)
    AI4CPS IHUB Grant (2024-26) ~ INR 1.06 Cr | Topic: Using Large Language Models to enhance learning efficiency and student engagement in Indian education system
    Main PI: Prof. Pawan Goyal
  5. SERB DST SRG (2021-23)
    SERB DST Startup Research Grant (2021-23) ~ INR 26 Lacs | Topic: "Learning from Rules and Data for Image Analytics"
  6. IIT Kharagpur FSRG (2022-24)
    IIT Kharagpur Faculty Startup Research Grant (2022-24) ~ INR 25 Lacs | Topic: The Role of Feedback in Vision-Language enabled Embodied Agents towards Applications in Desire Management
    Joint PI: Prof. Pawan Goyal
  7. Microsoft Accelerate Foundation Models Grant (May 2023 - Mar 2024)
    Microsoft: "Accelerate Foundation Models Academic Research" (2023) ~ USD 30000 | Topic: "Perils and Prevention of Prompt Injection Attacks for Large Language Models"
    Joint PI: Prof. Animesh Mukherjee
  8. Toloka AI Grant (2022)
    Toloka AI Annotation Grant (2022) ~ USD 300 | Topic: "Infusing Language Model with Affordances"

Codes and Datasets by the Group

  1. [2020 - 21] Crowdsourced and synthetic reasoning type-annotated data for NLI: TaxiNLI Dataset (2020) , LoNLI Dataset (2021)
  2. [2022] Semi-automated type-annotated Multilingual dataset for NLI: TaxiXNLI Dataset (2022)
  3. [2023] LogiGLUE: A Broad-coverage benchmark for Logical Reasoning with 10 In-domain and 15 Out-of-Domain datasets LogiGLUE Dataset (2023)
  4. [2023] LogiT5: A LLM Specialized in Logical Reasoning, fine-tuned on LogiGLUE LogiT5 Model (2023)
  5. [2024] MathSensei: A Tool-Augmented LLM for Mathematical Reasoning MathSensei Code and Data (2024)
  6. [2025] NLKI: A Lightweight Natural Language Knowledge Integration Framework for Improving Small VLMs in Commonsense VQA NLKI Code and models

FAQs for Propsective Students

  1. What are the projects that you will be working on?
    Here are some broad questions that we are interested in:
    • How do we reason transparently? What does it mean to reason transparently?
    • What are the complimentary reasoning abilities of neural and symbolic logical methods? How do you leverage that to build highly accurate and robust neuro-symbolic methods?
    • How do we reason with noisy and incomplete knowledge sources? Can we utilize semi-structured (or more natural) forms of knowledge representation?
    • How do we elicit missing knowledge?
    • How can human-in-the-loop help?
    See the project page and publication to see how solving such technical challenges may affect a range of applications -- from NLP, Vision, to automated mathematical reasoning.
  2. Is my expertise right for your lab?
    I am happy to welcome students with diverse expertise, ideas and background. But, based on my experience, following expertise seem absolutely essential. First, coding expertise (python mostly) and scripting knowledge (shell script, R is a plus). Being able to figure out your way through installation issues, knowledge about Git, shell scripting, and basic idea about GPUs are important. If you are an IIT-KGP student, make sure you have obtained A or above in PDS theory and Lab. Share a link to your Git repositories. Participation in popular open-source projects is a plus.
    Secondly, even if you are not familiar with ML, in-depth understanding of Algorithms, Probability and statistics is important. Knowledge of Logic, ML, DL, traditional NLP (parsing, POS tagging, chunking etc.), linguistics, or traditional image processing is a huge plus.
  3. How do I contact you?
    If you are an IIT KGP student, please send an inquiry to my email. Kindly mention which program you are enrolled in. If you are not a KGP student, kindly visit the IIT Kharagpur website to see how to enroll in the Ph.D or Masters program. I plan to take 1 or 2 external students who shows proven depth in at least one of the axes (linguistics/ML/DL/CV) through Publications/Courses/olympiads as interns (for one or two semesters). Feel free to email me if that interests you, and you fit the profile.