A
Abhyudya Bhatnagar
About
Detail
Bengaluru, Karnataka, India
Contact Abhyudya regarding:
Flexible work
Starting at
USD10/hour
Résumé
Jobs
verified_user
0% verified
- IGTM Engineering InternINFLXDJul 2025 - Current (9 months)Developed an end-to-end Prompt Generator backend using FastAPI + Supabase powering 500+ finance-domain prompts across VC, PE, Sell-Side, FP&A, and IR.Optimized fuzzy-search and metadata-ranking algorithms for semantic retrieval with under 10 ms query latency.Implemented Supabase data pipelines enabling query-based filtering, prompt generation, and multi-model analytics.Integrated PostHog dashboards tracking search behavior and lead-magnet performance, guiding SEO & roadmap.
- FGen AI InternFriska.AIMar 2025 - Jul 2025 (5 months)Engineered duplex Conversational AI pipeline (under 200 ms latency) integrating STT, ASR, and TTS.Architected a dual-LLM Q&A system boosting data throughput 5× and cutting manual work 80%.Trained domain-adapted LLMs on medical-nutrition datasets to deliver accurate, context-aware dietary recommendations.
- PDeveloper InternProcohatJun 2024 - Aug 2024 (3 months)Designed PowerBI dashboards for real-time insights, improving decision-making and UX.Migrated 7 years of physical data to a virtual datalake via SAP ABAP, increasing data accessibility by 25%.
Education
verified_user
0% verified
Bachelor of Technology in Computer ScienceIndian Institute of Information TechnologyNov 2022 - Current (3 years 5 months)
Projects (professional or personal)
verified_user
0% verified
- DDocuMind | PythonFeb 2025Developed an AI-powered document Q&A system using Retrieval-Augmented Generation (RAG) with Qdrant vector storage, reducing query latency by 400%.Integrated Google's Gemma2 27B model for accurate context-aware responses improving context-aware responses by 11% in internal benchmark testing and reducing average response time for complex queries by 21 milliseconds.
- KKnowledge Distillation Implementation | Python, JupyterJan 2025Achieved 94.4% accuracy by refining BERT into DistilBERT, reducing model size by 40% while retaining 97% performance.Implemented custom KL-divergence loss with temperature scaling to improve training efficiency.