Pranav Joshi

Machine Learning Engineer (M.Tech, IIT Jammu) spanning experience in applied ML, deep learning, and production GenAI/agentic systems. Currently building LLM orchestration, RAG, and multi-agent pipelines at Infocusp; previously spent 2.5 years as a Research Engineer at IIT Bombay on AI-driven cyber-defense and network anomaly detection. Track record of shipping systems that cut manual effort and turnaround time by 40–95%.

Machine Learning Engineer at Infocusp Innovations LLP (2024-01 – Present)

• Developed a Python-based experimental prototype to map ambient user context (screen data, audio, notifications) to proactive smart actions like calendar suggestions and agent calls.
• Synthesized test data pipelines using internal tools to rigorously evaluate and validate new action triggers prior to production.
• Refined prompt architectures through extensive iterative testing, to improve the accuracy of contextual suggestions, ensuring high-fidelity responses to ambient user signals.
• Engineered an end-to-end LLM pipeline that ingests a single research paper, maps its foundational lineage, and generates a comprehensive technical blog in under 4-5 minutes.
• Automated the retrieval of base-papers by parsing PDFs into Markdown (Docling, PyMuPDF4LLM) and querying OpenAlex and Semantic Scholar APIs via a multi-stage prompt sequence.
• Orchestrated a synthesis workflow to summarize cross-paper base concepts alongside the input paper, slashing manual research and reading time by over 95% (from 2+ hours down to 3 minutes per paper).
• Engineered a dynamic, stateful LLM orchestration system in Python, to drive multi-turn candidate surveys, cutting time-to-launch by 70% and eliminating 80% of manual creation effort.
• Developed modular OOP classes and complex prompt chaining to manage dialog state, dynamically generating context-aware follow-up questions aligned with specific user research criteria.
• Optimized GPT-4/4o prompt architectures—enforcing strict JSON schemas for deterministic database integration and utilizing Chain-of-Thought (CoT) summarization to reduce analysis turnaround time by 40%.

Research Engineer at IIT Bombay (2021-08 – 2024-01)

Contributed as IC in AI-enabled cyber defense research under Prof. Virendra Singh, replicating SOTA models on proprietary network security datasets.

• Architected a high-precision anomaly detection model analyzing large, complex network flow datasets to identify inconsistencies and fraudulent behavior patterns, achieving 99% classification accuracy.
• Explored and processed large-scale raw datasets, performing advanced feature selection and data cleaning to transform raw data into actionable statistical features for machine learning pipelines.
• Engineered a novel malware classification pipeline by transforming API function call sequences into NLP structures, fine-tuning BERT and Longformer architectures to achieve a highly precise Log-Loss of 0.037 on Microsoft datasets.

Research Intern in Deep Learning & Privacy Project at Tata Research Development and Design Centre (TRDDC) (2020-08 – 2020-10)

• Developed a text summarization pipeline for privacy policies, reducing reading time by 97% using deep learning-based extractive summarization techniques.

Trainee Software Engineer at Tibco Softwares (2016-07 – 2017-09)

• Designed and implemented scalable integration solutions using the TIBCO Suite (BusinessWorks, BusinessEvents), while orchestrating and observing deployments via TIBCO Hawk, Silver Fabric, and EMS.

Master of Technology in Computer Science – Indian Institute of Technology (IIT), Jammu (2019 – 2021)

Bachelor of Engineering in Computer Science – Shivaji University, Kolhapur (2012 – 2016)