Ameya Mhatre, software engineer
I build software that measures things. Agent pipelines that value IPOs, a model that prices NBA contracts against the cap, a screener that sizes risk before a trade, and audits that hold web apps to NIST CSF and ISO 27001.
Work
3 instruments
About
Most of what I build ends in a number someone has to trust. So the numbers come from code that can be tested, and the language model is only allowed to explain them.
Deterministic first
Ratios, cap math and stop distances are plain functions with tests. Models interpret; they never produce the figure.
Evaluate retrieval
At Mastek I benchmarked embedding strategies and Azure AI Search hybrid retrieval on real support tickets, so incident matching improved on evidence.
Audit what ships
I assessed a production web app against NIST CSF and ISO 27001. The checklists I wrote for it were reused on later engagements.
- Languages
- Python, JavaScript, SQL, C++
- Models and data
- scikit-learn, pandas, NumPy, XGBoost, Regression, Statistics
- Systems
- LangChain, LangGraph, FastAPI, React, Node.js, AWS, Git
- Security
- NIST CSF, ISO 27001, Threat modelling, Security audits
Record
2024 to 2028
B.Tech, Computer Science (Data Science)
Manipal Institute of Technology, Bengaluru
Aug 2024 to May 2028 (expected)
- CGPA 8.4 / 10
- Data structures and algorithms, probability and statistics, OOP, discrete maths
Intern
Mastek Ltd., Mumbai
Jun to Aug 2025; Jun to Jul 2026
- Evaluation pipeline for an AI incident-resolution system: embeddings and Azure AI Search hybrid retrieval on real support data.
- Multi-node LLM agents in LangChain and LangGraph, including a production blog-writing agent (research, draft, edit).
- Security assessment of a production web app against NIST CSF and ISO 27001.
Chair, IEEE Computational Intelligence Society
MIT Bengaluru student chapter
Aug 2025 to Aug 2026
- Led a chapter of 15+ members; workshops on AI and security for 200+ attendees.
- Budget, cross-team coordination and stakeholder communication.
| now, Oct 2026
- 8.4/ 10CGPA
- 15+chapter members led
- 200+workshop attendees


