AHMEDABAD, IN - 23.02°N 72.57°E©2026

Mann Ahalpara

AI/ML Researcher - Conversational AI · LLMs · Information Retrieval

Currently - Research Intern, Sentic Team @ NTU Singapore

Scroll to retrieve ↓
Large Language ModelsConversational AILong-Term Memory in AI AgentsRetrieval-Augmented Generation & Information RetrievalAgentic Systems & LLM ReasoningPersonalized & Affective AITrustworthy, Privacy-Preserving & On-Device AI

MEM_02/ About

Positioning / who is retrieving

About

I'm Mann Ahalpara, an AI/ML researcher working on a question that sounds simple and isn't: how can an LLM agent truly remember the people it talks to? As a Research Intern with Prof. Erik Cambria's Sentic Team at NTU Singapore, I study long-term memory for conversational agents: how they retain, retrieve and update what matters across many sessions, while staying private and consistent. I'm a final-year B.Tech ICT student at Adani University, ranked 1st in my department with a CGPA of 9.13/10. Beyond research, I build and deploy end-to-end AI systems, from citation-grounded RAG and explainable answer evaluation to ML pipelines and voice agents, and I've co-authored a paper at ICIDS 2025.

Indexed values

ICT Department rank
#1
CGPA / 10
9.13
LeetCode problems solved
600+

Education

  • B.Tech, Information & Communication Technology

    Adani University, Ahmedabad

    CourseworkMachine Learning · Artificial Intelligence · Analysis & Design of Algorithms · Data Mining & Business Intelligence · Discrete Mathematics & Graph Theory · Numerical Methods & Optimization · Database Management Systems · Data Structures · Signals and Systems

    2023 - 2027 (expected)CGPA 9.13/10 · Ranked 1st in the ICT Department
  • Class XII (HSC), Science

    GSHSEB

    2023Percentile 97.03

MEM_03/ Research

Now - what is being written to memory

Research

Now - Sentic Team, NTU Singapore

Research Intern (Remote)

Sentic Team, Nanyang Technological University (NTU)

Dates
May 2026 - Present
Place
Singapore, remote
People
Mentor: Tiesunlong Shen · PI: Prof. Erik Cambria
  1. 01

    Research privacy-preserving, long-term Emotional Support Conversation (ESC) systems: how agents retain relevant information and give consistent, personalized support across multi-session interactions.

  2. 02

    Investigate hierarchical and agentic memory architectures for storing, retrieving, updating and selectively using information across sessions.

  3. 03

    Experiment with memory-conditioned AFlow pipelines and Monte Carlo Tree Search (MCTS) on local language models to study contextual continuity and memory-guided reasoning.

  4. 04

    Explore on-device and privacy-preserving approaches for persistent agents; review work on personalized agents, affective computing and retrieval-augmented systems to find open research gaps.

Conceptual memory architectureSession n flows into an encoder, then a hierarchical memory store with episodic and semantic memory, then retrieval and update, then memory-guided reasoning with AFlow and MCTS, then a response. The store, retrieval and reasoning sit inside an on-device, private boundary.ON-DEVICE / PRIVATEUPDATE · FORGET01Session nMULTI-SESSION DIALOGUE02EncoderTURNS → REPRESENTATIONS03Hierarchical Memory Store04Retrieval & UpdateSTORE · RETRIEVE · UPDATE · FORGET05Memory-guided reasoningAFLOW · MCTS06ResponseCONSISTENT, PERSONALIZEDEPISODICSEMANTIC
Fig. 01 - Conceptual overviewNot a reported result

Research interests

  • Large Language Models
  • Conversational AI
  • Long-Term Memory in AI Agents
  • Retrieval-Augmented Generation & Information Retrieval
  • Agentic Systems & LLM Reasoning
  • Personalized & Affective AI
  • Trustworthy, Privacy-Preserving & On-Device AI

MEM_04/ Publication

Peer-reviewed / written to long-term storage

Publication

[1] ICIDS 20252025

“Analyzing Indian Stock Markets through Correlations: Comparative Insights”

Mann Ahalpara (Co-author). Proceedings of the International Conference on Infrastructure Development and Sustainability (ICIDS 2025), Atlantis Press. Guided by Dr. Ashlesha Bhise.

  • Comparative study across ten major Indian market sectors using Pearson, Spearman, Kendall's Tau and time-lagged cross-correlation.
  • Correlation analysis, network representations and comparative visualization of inter-sector dependencies and diversification opportunities.
  • Extended into CorrelationX, an interactive platform for multi-method correlation analysis.
Fig. 02 - 10 × 10 sector matrixIllustrative · random-seeded

Methods

Pearson · Spearman · Kendall's Tau · Time-lagged cross-correlation

MEM_05/ Work

Selected entries / built, deployed, cited

Work

04 / 04 entries retrieved

MEM_06/ Internships

Episodic memory / chronological

Internships

  1. May 2026-Present

    ENTRY 01

    Research Intern (Remote)

    Sentic Team, Nanyang Technological University (NTU) · Singapore

    Mentor: Tiesunlong Shen · PI: Prof. Erik Cambria

    • Research privacy-preserving, long-term Emotional Support Conversation (ESC) systems: how agents retain relevant information and give consistent, personalized support across multi-session interactions.
    • Investigate hierarchical and agentic memory architectures for storing, retrieving, updating and selectively using information across sessions.
    • Experiment with memory-conditioned AFlow pipelines and Monte Carlo Tree Search (MCTS) on local language models to study contextual continuity and memory-guided reasoning.
    • Explore on-device and privacy-preserving approaches for persistent agents; review work on personalized agents, affective computing and retrieval-augmented systems to find open research gaps.
  2. Jan 2026-Feb 2026

    ENTRY 02

    AI & ML Intern

    Edunet Foundation (AICTE × IBM SkillsBuild)

    • Built and deployed Study Buddy, an AI learning assistant (Gemini API + Streamlit) for explanations, summarization, quiz and flashcard generation, and PDF-based learning, using document-processing and prompt-engineering workflows.
  3. Jul 2025-Aug 2025

    ENTRY 03

    Product & UX Research Fellow

    Aspir

    • Six-week product sprint with a cross-functional startup team on AI-assisted entrepreneurship tools: AI-generated interface concepts, product workflows and early feature ideation.

MEM_07/ Stack

Semantic memory / tools in active use

Stack

  • Languages01

    • Python
    • SQL
    • Java
    • TypeScript/JavaScript
  • ML / DL02

    • PyTorch
    • JAX
    • scikit-learn
    • Transformers
    • Regression
    • Classification
    • Feature engineering
    • Model evaluation
    • Statistical analysis
  • LLMs & Agents03

    • Gemini
    • LLaMA
    • Ollama
    • Groq
    • Prompt engineering
    • AI agents
    • Conversational memory
    • MCTS
    • AFlow
  • Retrieval04

    • RAG
    • Sentence Transformers
    • Embeddings
    • Semantic/vector search
    • Reranking
    • Qdrant
  • Backend & Web05

    • FastAPI
    • Flask
    • Next.js
    • Streamlit
    • REST APIs
    • Auth & role-based workflows
  • Data06

    • PostgreSQL
    • MySQL
    • Supabase
    • Pandas
    • NumPy
    • NetworkX
    • OSMnx
  • Viz & BI07

    • Power BI
    • Tableau
    • Matplotlib
    • Seaborn
  • Ship08

    • Git
    • Docker
    • Vercel
    • Render
    • Streamlit Community Cloud

MEM_08/ Ask my memory

Retrieval / query this site

Ask my memory

memory://mann-ahalpara

Ask anything about Mann.

Try

MEM_09/ Achievements

Index / achievements & training

Achievements

  1. 01Ranked 1st in the B.Tech ICT cohortAdani University · CGPA 9.13/10
  2. 02National Technology Day 2024Represented Adani University at the event hosted by SAC-ISRO, Ahmedabad (selected on academic performance)
  3. 03600+ DSA problems solvedLeetCode
  4. 04Co-author, ICIDS 2025Atlantis Press
  5. 05Retrieval-Augmented GenerationNVIDIA Deep Learning Institute
  6. 06Hands-on Quantum Error CorrectionGoogle Quantum AI · stabilizer formalism, surface codes
  7. 07GenAI-Powered Data Analytics Job SimulationTata · Forage