Avinash P. Email & Phone Number

Pune, Maharashtra, India

I work on LLM and LLM Agents for use cases like chatbots, GUI generation, etc.I use Langchain, Llama-2 (chat), GPT, etc.Working on RAG and other fine-tuning approaches.

Leading a team of data scientists at TomTomFew of the projects that I am working on● POIs (Point of Interests) deduplication based on entity resolution - Trained deep learning-based entity matching/deduplication model with >97% accuracy for English speaking countries where each country has more than a million entities - Used blocking technique to reduce number of computationally expensive computations (time reduction by a factor for 10X) - Adopted multi-language models for non-English speaking countries (Experiment with BLOOM language model) - Working on embedding size reduction - Sentence Transformers, Autoencoder etc.● Improved entities name matching - Improved name matching by 5% using Sentence Transformers (BERT based uncased)● Search Ranking Signals - Analyzed footfall data for places in Netherlands - Created a signal (confidence score) which is one of the parameters for deciding the ranking of place search in TomTom map● Metric for Routing Points (Routing points make correlation between places and streets on TomTom maps) - Created a metric for RP on new TomTom maps by comparing with older version - The extension of same project is creating metric considering ground truth data

Pune, Maharashtra, India

- Propensity Models like Likelihood to Buy, Likelihood to Convert, Likelihood to pay full price for retail with 95% score accuracy, worked on data of more than 3 years. Millions of data points per company.- Reinforcement Learning model for send time optimisation- Worked on End-to-End ML pipeline- Skills: Python, Spark, Qubole, Machine Learning, Reinforcement Learning, Tensorflow, AWS, Reinforcement Learning

Leading ML on code with the use of various Machine Learning techniques.- Code fix recommendation- Designed model architectures based on SOTA language models and Graph NNs to perform code analysis.- Constructed datasets to benchmark code analysis tasks such as bug localization & code recommendation.- Designed CoFEx (Code Feature Extraction) model and managed its integration with the Gamma Recommendation Engine- Managed the model training for downstream tasks such as code auto-completion, bug detection and localization.- Worked on TFIDF, LSTMs, Seq2Seq, RNNs, and transformer models (BERT and XLNET)

Pune Area, India

Working on machine learning-based static analysis of code.

Pune Area, India

- Worked on occupancy calculation considering resources like threads, register files, shared memory, constant bank, texture memory to determine if Compiler has produced optimal code- Developed a framework for frame-based GPU performance/functionality verification- Verified synchronization done in Nvidia Compiler using assembly language code- Texture/Surfaces verification and framework development in CUDA/OpenCL for the same- CUDA Memory Consistency Model Verification (which is based on C++ 11 memory model)- Verified all the instructions added to PTX assembly for all the architectures from Kepler onwards- Wrote a framework for assembler verification from scratch- Wrote optimal tests in Assembly language (closest assembly language to Nvidia GPU which is directly translated to GPU binary)- Mentored an intern for project “Generating automatic tests for Compiler synchronization- verification” in Python; we generated 45 tests from 8 seed tests with isomorphic graph approach and generalised it later to create huge automated test corpus

Pune, Maharashtra

Cs-441 Applied Machine Learning

Cs-447 Natural Language Processing

Cs-410 Text Information Systems

Computer Engineering, Computer Science

Data Structures, OS, Advanced Algorithms, Networking etc.