Hi, my name is
Ashmit Mandal.
I build intelligent, low-level systems.
I’m a computer science undergraduate specializing in Generative AI and systems engineering. Currently, I serve as Head of Research at K-1000 and lead game development architectures with AI agents at AISOC.
About Me
Hello! My name is Ashmit and I enjoy building secure, performant systems and dynamic AI experiences. My journey in computer science began with deep diving into algorithms and competitive programming, which sparked my passion for operating systems kernel optimizations and neural network engineering.
Fast-forward to today, and I’ve had the privilege of serving as the Head of Research at K-1000, guiding dozens of students in drafting international manuscripts, and leading the Game Development domain at AISOC inside KIIT University. My core interests lie at the intersection of systems optimizations, explainable AI, and large-scale application design.
I am also active in publishing scientific research. I’ve recently co-authored papers featured in Springer (QI-Net for Explainable AI) and IEEE (efficiency trade-offs of model pruning).
Here are a few technologies I’ve been working with recently:
- C / C++
- Java
- Python
- Dart / Flutter
- TensorFlow / PyTorch
- GCP / Supabase
- Docker
- Systems Programming
Where I’ve Worked
Summer Intern @ NTPC Limited
Jun 2026
- Developed a CCTV analysis pipeline automating PPE detection for NTPC Limited, achieving significant advancements in workplace safety.
- Utilized RTSP protocols to create a fully automated system for protective detail analytics.
- Enhanced operational efficiency by eliminating the requirement for human observers.
- Achieved an 85% detection accuracy against live-testing validation footage from NTPC Vindhyachal upon deployment on the internal NTPC network.
Some Things I’ve Built
Featured Project
Explainable Credit Risk (ECR) Engine
Architected an explainability-first credit underwriting engine for India's informal economy. Replaces black-box risk assessment by leveraging alternative data (UPI/SMS) to compute fair, RBI-compliant lending decisions. Implemented row-level database encryption, monotonic fairness constraints, SHAP reason code explanations, and DiCE counterfactuals generating actionable approval paths.
- FastAPI
- Next.js
- Supabase
- LightGBM
- SHAP
- DiCE
![Explainable Credit Risk (ECR) Engine]()
Featured Project
KIIT-Assist
Campus utility application supporting student, admin, and staff roles. Implemented advanced geofencing using a custom Ray-Casting algorithm to validate user location boundaries across 20+ campuses. Engineered a secure Supabase database structure relying on Row Level Security (RLS) to safeguard student complaints.
- Flutter
- Supabase
- Geofencing
- Row Level Security (RLS)
![KIIT-Assist]()
Featured Project
Queen's Gambit: N-Queens Solver
Developed an interactive Flutter mobile application to solve and visualize the classic N-Queens chess puzzle. Engineered an optimized backtracking algorithm customized with interactive play modes, real-time threat-level coordinates highlighting, and persistent puzzle states powered by
flutter_blocarchitecture for a smooth 60 FPS user experience.- Flutter
- Dart
- BLoC
- Backtracking
![Queen's Gambit: N-Queens Solver]()
Other Noteworthy Projects
view the archiveGradient and Reason
A personal blog bridging machine learning and philosophy. Essays on structural causal models, latent space geometry, interpretability, and the epistemology of neural networks.
Flutter 8-Puzzle Solver
Architected a high-performance mobile solver game using Clean Architecture and BLoC state management to maintain a fluid 60 FPS user experience. Integrated the A* Search algorithm using Manhattan heuristics, computing optimal solution paths in under 50ms.
Voice RAG Assistant (Murf.ai Runner-Up)
Runner-Up of Murf.ai Challenge 4. Built a low-latency voice-to-voice RAG assistant with domain-specific knowledge bases, high-performance FAISS matching, and synthesized speech output.
Model Pruning for Low-Powered AI (IEEE)
Analyzed the computational efficiency trade-offs of deep learning compression techniques (pruning & quantization) for high-performance deployment on edge and low-power devices.
Interpretable Neural Networks (Springer)
Authored research introducing QI-Net, an interpretable-by-design CNN achieving 97.2% Malaria detection accuracy while maintaining intrinsic explainability via orthogonal convolutions.
Hands-Free Agentic AI (Murf.ai Winner)
Winner of Murf.ai Challenge 3. Developed a hands-free Agentic AI automation workflow leveraging LLM agent reasoning, voice input, and automated action execution.
What’s Next?
Get In Touch
I am currently open to new career opportunities, research collaborations, and engineering challenges. Whether you have a question, a research idea, or just want to connect, feel free to reach out!
Say Hello