UC Berkeley · B.S. Computer Science & B.S. Data Science · Class of 2027

Aahan Bagga

I build ML systems that hold up in production.

Currently raising anomaly-detection recall at Pronto Networks. Previously: agentic RCA, graph-based routing engines, and LLM orchestration pipelines. Looking for ML, Data Science, and Software Engineering internships for Summer 2027.

Download résumé See projects

01 — About

I like the layer where raw data meets something that actually has to work: an anomaly detector that catches a real outage, a routing engine that ships a real route, an agent system that doesn't fall over under load. Most of my work sits across that line — part modeling, part systems engineering, part "does this survive contact with production." Outside of that, I co-founded a nonprofit that's taught data science to 1,000+ students across three continents, which is where I actually learned how to explain any of this clearly.

3.6 GPA / 4.0
1.1% Top admit percentile, UCB
4 ML / SWE internships
2027 Graduating

02 — Experience

Where I've worked

  1. Machine Learning Intern

    Pronto Networks · Walnut Creek, CA · May 2026 – Present
    • Built a heterogeneous graph neural network (HGNN) anomaly-detection pipeline — site-level models feeding an org-level aggregator — raising recall from 40% to 93% while cutting false positives 50%+.
    • Designed a two-stage root-cause-analysis pipeline that isolates the responsible device and hands off LLM-driven, agentic remediation steps.

  2. Software Consultant Contract

    IBM · Berkeley, CA · Jan 2026 – May 2026
    • Engineered an autonomous TORCS driver in Python using IBM Granite for real-time vehicle control and telemetry, cutting lap times from 60+s to 28s on the Corkscrew Circuit for IBM's AI Racing League.
    • Tuned steer gain, target speed, and braking logic against live telemetry to push circuit performance.
    • Led a multidisciplinary team through IBM's Global Progression Funnel, a tiered international competition.

  3. Machine Learning Researcher

    HERO2 · Berkeley, CA · Jan 2026 – May 2026
    • Built a multimodal navigation engine using HA* and Dijkstra's UCS to optimize routes across time, cost, and energy constraints.
    • Cut Neo4j graph query runtime 96% (50s+ → <2s), enabling real-time pathfinding at scale.
    • Architected a routing system layering real-time transit APIs over graph abstractions for low-latency suggestions.

  4. Data Science Intern

    CerebrAIx Technologies · San Francisco, CA · Feb 2025 – Sep 2025
    • Orchestrated Xpredict, a multi-model inference pipeline combining GPT-4, Claude, and Gemini with SBERT to score candidate-job alignment via high-dimensional semantic similarity.
    • Built a RESTful API around a hybrid cosine-similarity scoring architecture for fitment metrics.
    • Managed PDF/DOCX data ingestion and benchmarked 5+ models on quality, cost, and latency.

  5. Co-Founder & Director, Data Science Camp

    Kids Teach Tech · San Francisco, CA · Sep 2020 – Dec 2024
    • Co-founded a nonprofit delivering free data science and AI classes to underserved youth across the Bay Area and internationally.
    • Grew the team from 5 to 100+ members; helped win the 2024 East Bay Economic Development Alliance Innovation Award for Education.
    • Directed 40+ instructors running a 300+ student camp at UC Berkeley, in partnership with UCB IEOR and Georgia Tech ISYE.

03 — Projects

Selected builds

A mix of production ML, graph algorithms, and infrastructure — some built on internships, some on my own time.

HGNN Anomaly Detection & RCA

40% → 93% recall

Site-level heterogeneous graph models feeding an org-wide aggregator, paired with a two-stage root-cause pipeline that names the failing device and drafts agentic remediation steps.

PyTorchGNNsLLM AgentsAnomaly Detection
Employer IP — code available on request

Multimodal Navigation Engine

96% faster queries

Routing engine combining HA* and Dijkstra's UCS to optimize across time, cost, and energy, layered over real-time transit APIs. Cut Neo4j query time from 50s+ to under 2s.

Graph AlgorithmsNeo4jTransit APIsPython
Research project — code available on request

AgentLedger

28 companies analyzed

B2B SaaS proxy that enforces real-time cost budgets, kill switches, and anomaly detection across production AI-agent workflows, built after a competitive gap analysis in agent observability.

Node.jsFastAPIRedisDocker
GitHub ↗

Xpredict

3-model ensemble

Multi-model inference pipeline blending GPT-4, Claude, and Gemini with SBERT embeddings to score candidate-job fit through recursive prompt engineering and hybrid cosine-similarity scoring.

LLM OrchestrationSBERTREST API
Employer IP — code available on request

W2CAI

7k+ structured entries

Full-stack AI product-discovery platform for Asian fashion — parallel scraping and LLM-based extraction pipelines feeding automated sourcing with integrated reviews.

Next.jsReactLLM Extraction
GitHub ↗

Autonomous TORCS Driver

60s → 28s lap time

Real-time autonomous racing agent built on IBM Granite for IBM's AI Racing League, tuning steer gain, target speed, and braking logic live against telemetry.

PythonIBM GraniteReal-time Control
Competition project — code available on request

04 — Skills

Toolbox

Languages

PythonJavaSQLCRGo

ML / AI & Data

PyTorchTensorFlowScikit-LearnPandas NumPyMatplotlibLangChainTransformers RAGAgentic AIA/B TestingEDA

Systems & Backend

DockerGit / GitHubREST APIsFastAPI Express.jsRedisMLOps

05 — Education

University of California, Berkeley

Class of 2027

B.S. Computer Science & B.S. Data Science — Early Admit, top 1.1% of applicants · GPA 3.6/4.0

Data Structures & AlgorithmsMachine Learning Agents & Generative AIDatabase Systems SICPArtificial Intelligence Data Science Principles & TechniquesComputer Security Linear AlgebraMultivariable Calculus
Copied to clipboard