Applied AI Engineer with an M.S. in Statistics who builds production retrieval and agent systems and the evaluation infrastructure that keeps them honest. Built an evidence-grounded RAG and knowledge-graph platform over ~1,600 civic meetings with claim-level citation validation, abstention on insufficient evidence, and a versioned evaluation harness covering retrieval, grounding, attribution, conversation and security behavior. Also built a canonical public-record data platform spanning 1.7M+ profiles, and works below the API layer on GPU memory, quantization and local inference performance.
Technical Skills
Experience
OpGov.ai
AI Engineer → Founding AI Engineer (Aug 2025)
January 2025 - Present
San Francisco Bay Area
- Built an evidence-grounded RAG and knowledge-graph system over ~1,600 civic meetings, combining BM25, dense retrieval and Neo4j graph traversal through Reciprocal Rank Fusion, CrossEncoder reranking and parent-context expansion
- Implemented constrained generation against an approved evidence set with deterministic claim and citation validation, so unsupported factual claims and invented citations are rejected before reaching a user
- Developed a versioned evaluation harness measuring Recall@5/10, MRR, meeting and speaker attribution accuracy, citation coverage and correctness, abstention behavior and prompt-injection resistance
- Engineered a canonical MongoDB platform unifying 1.7M+ public-record profiles and ~4M historical participation records by normalizing incompatible county schemas
- Reconciled a 734,457-record snapshot using typed identifiers and demographic matching, identifying 20,060 new profiles while isolating a single ambiguous identity for review rather than forcing an incorrect merge
- Added layered guardrails covering prompt injection, isolation of untrusted retrieved content and secret detection, with authenticated persistent chat sessions and full request tracing
Petpin AI
February 2024 - August 2024
San Francisco, CA
- Developed Python pipelines for real-time ingestion, preprocessing, TensorFlow Lite inference and API communication under device memory, battery and latency constraints
- Built React dashboards visualizing live telemetry, behavioral signals, device health and model outputs from connected hardware
- Connected on-device inference to backend services and user-facing interfaces, creating an end-to-end path from raw sensor data to interpretable product information
- Created monitoring and diagnostic tooling that helped hardware, software and product teams investigate field issues
R P Era
August 2018 - August 2023
India
- Founded and ran an engineering and 3D-printing services business, delivering 100+ customer projects across engineering and dental applications from discovery through acceptance
- Translated vague customer requirements into technical specifications, material decisions, production plans, estimates and finished components
- Built a Python and Flask order-management system covering quotations, job tickets, inventory, invoicing, production status and delivery workflows
- Automated repetitive CAD-file preparation for laser engraving and 3D-printing workflows, reducing manual handling and preventable production errors
CSU East Bay
August 2024 - December 2024
Hayward, CA
- Taught probability, statistical inference, hypothesis testing and sampling to a cohort of 22 students through structured lessons and applied examples
- Designed lectures, exercises and assessments connecting statistical concepts to real analytical questions
- Explained quantitative methods to students with varied mathematical backgrounds, strengthening communication with non-specialist audiences
Key Projects
Hybrid and graph RAG over ~1,600 civic meetings with versioned idempotent ingestion, BM25 + dense + graph retrieval, RRF fusion and CrossEncoder reranking.
Impact: Evidence-sufficiency gating + claim-level citation validation
Versioned harness covering Recall@5/10, MRR, attribution accuracy, citation coverage/correctness and abstention, with manually verified golden datasets and hard negatives.
Impact: Used as a release gate across pipeline versions
Multimodal autonomous agent system for complex task automation, built against the Model Context Protocol and agent-to-agent communication.
Impact: 3rd Prize
Six experiments on local LLM inference, GPU memory and transfer behavior, scheduling policies, camera-to-GPU latency, and QLoRA fine-tuning.
Impact: Tokens/sec, VRAM, and latency measured — not assumed
Unified 1.7M+ profiles and ~4M historical records from incompatible county schemas into canonical profile, election and longitudinal-history models.
Impact: 20,060 new profiles reconciled, 1 ambiguous identity isolated
Next.js and MongoDB geospatial app with GeoJSON, 2dsphere indexing and live device location, surfacing the nearest eligible records with contact actions.
Impact: Live nearest-record search with real-time location
Education
California State University, East Bay
2023 - 2025
GPA: 3.9 · Heebok Park Scholarship
Data Science concentration with focus on statistical modeling, machine learning, and applied analytics.
University of Mumbai
2014 - 2018
Awards
3rd Prize — MCP × A2A Hackathon
Built a multimodal autonomous agent system for complex task automation, using the Model Context Protocol (MCP) and agent-to-agent (A2A) communication.
Heebok Park Scholarship
Awarded during the M.S. Statistics program at CSU East Bay.