I design and build end-to-end AI systems that operate under uncertainty — focusing on evaluation, reliability, and real-world constraints rather than demos or toy benchmarks.

I work at the intersection of AI research and production engineering, with a focus on systems that must:

• Handle noisy, imperfect real-world inputs
• Produce deterministic, explainable outputs
• Be evaluated, measured, and stress-tested — not just generated

My work emphasizes architecture, evaluation, and failure modes over surface-level demos.

• Layout-aware OpticalRAG system for legal & procurement documents
• OCR → layout parsing → retrieval → LLM reasoning → deterministic risk scoring
• Designed for noisy PDFs, ambiguous clauses, and explainable outputs
• Includes tests, failure handling, metrics, and end-to-end execution artifacts

Focus: document understanding, evaluation under noise, system reliability

• Agent-based interview simulation and evaluation pipeline
• Generates structured, rubric-driven feedback (clarity, confidence, correctness)
• Emphasizes evaluation logic, scoring consistency, and failure analysis
• Produces machine-readable reports from live or scripted inputs

Focus: human evaluation, subjective signal modeling, scoring systems

• Real-time traffic intelligence pipeline
• Detection → tracking → violation logic with latency, FPS, and memory profiling
• Built with performance constraints and system metrics as first-class concerns

Focus: edge vision, performance engineering, real-time systems

• Pragya-OS: Exploration of agent-native operating systems for long-horizon planning, memory, and reflection
• Animark-AI: Experimental multimodal generation pipelines for visual storytelling and advertising

(Research explorations — not positioned as finished products)

Instead of videos, my projects emphasize reproducible proof:

• End-to-end input → output examples
• Deterministic JSON / report outputs
• Latency and pipeline metrics
• Explicit failure cases and trade-off documentation

These artifacts are available inside each repository under /examples, /docs, and /assets.

Modeling & Research
Python • PyTorch • TensorFlow • Scikit-learn • OpenCV • Pandas

Systems & MLOps
Docker • Linux • AWS • GCP • Git • MongoDB • MySQL

I write about deep learning fundamentals, optimization, and applied AI:

• CNNs and modern computer vision
• Optimizers, normalization, and training dynamics
• Practical trade-offs in real-world ML systems

📖 Articles published on Artificial Intelligence in Plain English and Python’s Gurus.

• Applied ML / AI Systems Engineering roles
• Evaluation-driven AI and agentic systems
• Multimodal intelligence under real-world constraints

If you’re interested in how systems behave when things go wrong, we’ll get along well.