Sillah B.-AI SoC Trainee

Skills: Python, LLMs, Multi-Agent Systems, RTL, SystemVerilog

• Developed a multi-agent system for automated RTL generation, coordinating specialized agents across design, verification, and refinement stages.
• Fine-tuned Llama 3.3 on SystemVerilog code generation tasks, improving syntactic correctness and adherence to hardware design conventions.
• Explored memory architectures and techniques for multi-agent systems, enabling context retention and improved coordination across iterative RTL generation cycles.
• Investigated agentic design patterns and inter-agent communication strategies to improve reliability and reduce hallucinated or non-synthesizable RTL output.

Skills: Python, AI, GCP, automations, pipelines, monitoring, data cleaning, Docker, Postgres, BigQuery

• Designed and implemented an intelligent email generation agent for highly customized cold outreach, leveraging MCP tool calling, LangGraph, and LangChain to evaluate outputs and continuously refine email quality.
• Created an asynchronous, authenticated MCP server using the FastMCP framework, exposing PostgreSQL and Elasticsearch through secured endpoints via Google OAuth, and persisting encoded tokens in Redis to ensure secure, authenticated access to internal data structures and persistent authentication.
• Designed scalable vector retrieval systems using Postgres (pgvector) for product–customer tag matching and Qdrant for large-scale embedding storage and recommendation of projects and products.
• Built scalable AI pipelines using embeddings for vendor and product recommendations based on customer profiles and structured tagging.
• Deployed Kestra as a pipeline orchestrator on a Kubernetes cluster, configuring fault-tolerant data extraction jobs on pre-emptible pods and integrating Pub/Sub for asynchronous messaging.
• Led development of internal tool to generate RFP signals and automate sales outreach, integrating seamlessly with HubSpot.
• Created AI-driven features for leads recommendation and targeting based on client metadata.
• Built data pipelines using GPT-4.1, Claude, and DeepSeek to extract structured data from spend reports.
• Fine-tuned LLMs for improved accuracy in purchase and domain-specific data extraction tasks.
• Integrated CI/CD workflows with Grafana dashboards for real-time monitoring and alerting.
• Developed a graph-based knowledge system using embeddings for semantic chatbot document retrieval.
• Designed data validation and sanity checks for LLM-based data extraction workflows.
• Handling processing and loading of tables reaching upto 148M+ rows of structured data and creating sync pipelines for them.

Skills: C++, python, computer vision, mathematical techniques(linear algebra, calculus) for problem solving, camera calibrations, jenkins, kubernetes, Docker

• spearheaded the development of an internal tool to gauge sequence complexity, annotation counts, time spent and revisions for different modules [object, pedestrian, traffic sign detection etc] made to track annotator and QA performance of a team of 200 annotators.
• Extracting data and creating Kpis in C++ for evaluation of models concerning self driving cars for Audi AG, Cariad and AAI.
• Overlooking annotation and labelling specifications and suggesting plausible changes required for object and pedestrian detection models
• Applying AI algorithms as per need such as Hungarian algorithm for assigning ground truth to perception generated by the model.
• Maintaining the post processing pipeline in python for interpolations, distance and angle estimations from lidar data.
• Assisted in migration of cloud to aws by setting up kubernetes cloud, job dsls, seed jobs and jenkins for 120 Scala and C++ pipelines.

• Learning and training YOLO-v3, v4, v5 on hospital prescriptions to coordinate text with headings such as dosage, medicines, date, etc., and being able to digitally extract and store such records
• Deploying the application using Flask.
• Led to gaining expertise in pandas, PyTorch, matplotlib, Python, machine learning, and deep learning concepts.