Software Engineer & AI Researcher specialising in LLM infrastructure, multi-agent frameworks, and deterministic code generation. Currently a first-year undergraduate at VIT Bhopal University, pursuing B.Tech in Computer Science & Engineering (AI & ML).
My work is focused on one question: can we make small, local language models reliable enough to replace expensive cloud APIs in real engineering workflows? I think the answer is yes — but only if we stop treating verification as an afterthought.
Essays
Experience
- Accelerated video preprocessing 53× (4+ hours to 4m 32s) on an 11 GB, 60fps 4K source on CPU-only infrastructure by replacing OpenCV random-seek decoding with a sequential I-frame-only FFmpeg pipe, eliminating 99.7% of redundant decodes and removing GPU dependency.
- Boosted pipeline throughput to 119 fps (4.74× realtime) in a 2-vCPU environment by replacing a MediaPipe face detector with an ONNX-runtime YuNet model and streaming raw pixel buffers directly into inference, eliminating intermediate disk I/O.
- Decreased fragmented speaker tracks from 1,527 to 7 via a Union-Find spatial merge algorithm atop a greedy IoU tracker, reducing downstream per-clip crop processing overhead by >99% while improving active-speaker detection accuracy.
- Engineered zero-loss fallback infrastructure for degenerate video scenes with automated center-crop synthesis and metadata tagging when faces are absent, validated across a 23/23 automated regression test suite.
Projects
Research
Currently
Lead of the AI/ML & Innovation Team at Gen AI Community, VIT Bhopal. First year of a four-year undergraduate degree. Building in public.
Interested in: local LLM inference, deterministic verification of probabilistic systems, multi-agent orchestration, and the gap between benchmark performance and real-world reliability.