Madhav Gupta

As a Full-Stack AI Intern, Madhav developed a secure RAG-based Q&A system for financial documents, integrating a full-stack pipeline and implementing guardrails for accurate response evaluation.

• Developed a Proof-of-Concept (POC) RAG-based Q&A system for financial documents using Ollama & Qdrant, efficiently handling PDFs with 10+ pages.
• Integrated an end-to-end pipeline with FastAPI, user authentication, and an enhanced Angular UI to enable secure PDF querying.
• Explored and applied the RAGAS framework to implement guardrails and enhance context matching for accurate response evaluation.

As an AI Research Intern, Madhav conducted advanced computer vision research, significantly improving detection accuracy and reducing false positives across surveillance systems.

• Boosted loitering detection by 50% using contrastive learning for background-invariant clustering across 50+ cameras.
• Fine-tuned a Swin Transformer with K-fold cross-validation, achieving 83% accuracy in image segregation.
• Reduced false positives by 80% across 100+ cameras through 2D linear regression for height-vs-(X,Y) coordinate filtering.
• Received a Letter of Recommendation for outstanding contributions to real-world computer vision problems.

As a Generative AI Solutions Intern, Madhav enhanced information search efficiency by integrating generative AI solutions and secured leadership approval for future scaled product development.

• Integrated 4 applications via APIs, SharePoint, and Copilot Studio, boosting information search efficiency by 3x across platforms.
• Implemented robust API connectivity with topic triggers to fetch relevant information and generate RAG-style answers.
• Presented a successful Proof-of-Concept (POC) to leadership, securing approval for future scaled product development targeting 10K+ users.
• Awarded a pre-placement interview, recognizing outstanding contributions and potential for a full-time return offer.

As a Conversational AI and LLM Intern, Madhav developed an end-to-end conversational analytics pipeline and fine-tuned a GPT-2 summarizer model for customer conversations.

• Engineered an end-to-end conversational analytics pipeline encompassing data preparation, translation, fine-tuning, and inference.
• Created 200+ conversations to identify pain points, agent scores, and semantics for GPT-2 fine-tuning and generation.
• Translated OpenOrca and Samsum datasets using Opus and fine-tuned a GPT-2 summarizer model with SimpleTransformers to generate concise summaries of customer conversations in Tagalish.

Mentored participants in developing a Computer Vision-based Sudoku Solver during the WiDS Bootcamp.

• Mentored 10+ participants in the WiDS Bootcamp on developing a Sudoku Solver using Computer Vision.

Managed top-tier industry relations and enhanced institutional bonding during Impulse, EESA's annual festival.

• Served as the Point of Contact (POC) for top management from companies like Samsung, Mercedes, and Reliance Jio.
• Fostered valuable connections and enhanced institutional bonding through interactions with industry leaders.

Coordinated the placement season at IIT Bombay, facilitating interviews for thousands of students and managing firm relations.

• Coordinated a team of 250+ members to facilitate interviews for 2000+ students during the 2022-23 placement season.
• Assisted 20+ firms in conducting interviews, handling student queries, and maintaining interview records.

Led a thesis project focused on enhancing vision model interpretability for Out-of-Distribution (OOD) data, developing a novel submodular selection and perturbation-based uncertainty framework.

Designed and implemented a Bi-LSTM and transformer-based framework to improve NLP model interpretability and performance on sentiment analysis and text generation tasks.

Conducted comprehensive statistical and predictive analysis on large datasets, including NEET-UG scores and food features, to identify anomalies and predict trends.

Developed an AI-powered system for transforming tender management through semantic search, full lifecycle automation, and automated scraping.

Designed and implemented a multi-agent system to automate AI workflows from business-specific inputs and orchestrate subtasks dynamically.

Evaluated and optimized DETR-based object detection models by fine-tuning backbones and layers to enhance accuracy and efficiency.

Developed an AI-powered resume assistant by fine-tuning a Mistral-7B model and optimizing it for efficient inference.

Part of a 70+ member team designing driverless electric race cars for Formula Student competitions, representing the first Indian contingent at FS-AI 2023 at Silverstone Circuit.

Participated in an algorithmic trading hackathon, classifying stocks, predicting index returns, and implementing an optimized portfolio allocation strategy.

Achieved 3rd place in the Electrical Engineering department-wide table tennis tournament.

Secured 2nd position in a competitive algorithmic trading hackathon.

Ranked among top 15 of 250+ teams in Finsearch, presented to industry professionals, and won a cash prize.

Achieved 99.989 percentile and All India Rank 164 in JEE Main examination amongst 1 million+ students.

Secured All India Rank 724 out of 0.15 million+ candidates who appeared for JEE Advanced examination.

Recipient of the prestigious KVPY Fellowship, recognizing academic excellence in science.

Recipient of the prestigious National Talent Search Examination (NTSE) scholarship, ranked 3rd in Stage-1.

Qualified for Regional Maths Olympiad (RMO) by securing a rank in Top 1%ile in Pre-RMO.

Co-authored a research paper on developing a benchmark for face and license plate recognition in low-resolution videos, submitted to a top-tier AI conference.