AI engineer in New Delhi. Agents, retrieval and computer vision, and the evaluations that keep them honest. Recently completed AI engineering internships at Think Decor and FlyRank AI.
View work01 About
I started with CNNs and captioning models. Now I build multi-agent pipelines and retrieval systems, and I spend just as long measuring them. I report numbers with their denominators, and caveats with their numbers.
02 Experience
Built computer vision and AI features for Think Decor, an interior design company.
Selected for the FlyRank AI internship programme. Built backend AI services and APIs in Python: LLM-powered applications, API design and testing. Finished with a certificate of completion.
CGPA 9.23 / 10. Machine learning, deep learning and data systems.
03 Selected work
A test suite for agents that can promise money.
A policy-conformance harness and deploy gate. Every failing row carries the policy clause and the contradicting sentence, both verified as literal substrings.
Reported as a lower bound; limitations documented in the repo.
An adaptive adversarial fraud lab.
Breeds synthetic fraud behaviours that target a detector's blind spots, maps them as a 120-cell atlas, hardens the detector, then tests transfer to an unseen attack family at a fixed false-positive budget.
Simulator-only; fidelity is calibrated to the simulator, not real card traffic.
Dashboards from spreadsheets, answers from documents.
A LangGraph router sends each question to SQL, sandboxed statistics or grounded retrieval, with OCR for scanned pages and Groq-to-Mistral failover.
Fourteen more projects, from agents to audio.
See the index04 Open source
Compatibility fixes, input validation, flaky tests and missing coverage. The unglamorous kind, which is how I learn a codebase.
05 Activity
Public GitHub contributions over the last twelve months, loaded live on every visit.
Loading contributions…
06 Toolkit
07 Contact
Available for internships · usually replies within a day
Looking for AI/ML engineering and research internships, remote or in India. If you're working on agents, evaluation or applied vision, I'd love to hear about it.
AI answers from Ashish's portfolio. It can make mistakes.