Program Leadership · AI Research
AI program leader, building real systems and publishing the research.
I turn ambiguity into shipped products and peer-reviewed research — across AI, platforms, and emerging tech. Outcome-driven and calmly decisive.
7 published AI papers (arXiv, NeurIPS 2026, TMLR) on LLM systems — including Agaram 2.5B, a Tamil–English model trained from scratch with open weights.
Shipped production AI: a live storytelling platform and an offline, on-device clinical-triage assistant.
Led $25M+ multi-state SaaS programs across 7 jurisdictions; 99.99% uptime migrating to AWS.
Experience
AI research, shipped products, and enterprise delivery.
AMD
Firmware and AI Program Manager
Sep 2026 — Present
Austin, TX
- Manage the AI program within the firmware organization of the Data Center GPU business unit.
Murai Labs
Founder & AI Researcher
2025 — Present
Austin, TX
- Authored seven AI research papers (arXiv, NeurIPS 2026, TMLR) on large-language-model systems — spanning direct Apple Neural Engine programming, cooperative model training, multi-model orchestration, evolutionary mixture-of-LoRA fine-tuning, and pre-registered cheap-baseline screening for evolutionary methods.
- Shipped production AI products end-to-end: StoryForge Kids, a live iOS + web personalized-storytelling platform, and Marunthagam, a Tamil-first clinical-triage assistant running a 5GB language model fully offline on an Android phone (0/12 missed emergencies, 100% safety-refusal on adversarial prompts).
- Delivered headline research results — +21.76% cross-lingual model-fusion gain (KALAVAI) and 8.5× Apple Neural Engine compilation speedup (Orion) — each reproducible from committed code and audit-tracked artifacts.
Procore Technologies
Enterprise Project Manager
Dec 2025 — Sep 2026
Austin, TX (Remote)
- Managed project delivery for enterprise public sector and government clients.
- Led cross-functional coordination across engineering, product, customer success, and compliance.
- Drove change management using SDLC methodology — zero disruption to government operations.
Selected Work
High-clarity case studies.
Agaram 2.5B
Tamil–English language model trained from scratch: its own tokenizer, a 50B-token pretraining run, chat fine-tuning and tool calling. Open weights under Apache-2.0.
↳ Pretrained for about USD 4,765 on one 8×H100 node; translates Tamil–English better than Sarvam-1 in all six directions measured.
StoryForge Kids
Live personalized children's storytelling platform (iOS + web) with a multi-pass story engine and persistent character memory.
↳ Shipped end-to-end across iOS and web — a production AI product with real users, not a prototype.
Marunthagam
Tamil-first clinical-triage assistant running a 5GB language model fully offline on an Android phone for rural community-health workers.
↳ 0/12 missed emergencies; 100% safety-refusal on adversarial prompts. Submitted to the Gemma hackathon.
PINNAL
RNA-therapeutic candidate ranker that prioritizes antisense-oligo interventions for splice-altering variants using an ensemble of genomic foundation models (Evo 2, Borzoi, Enformer, ViennaRNA).
↳ Pre-alpha, demo-ready. Surfaced all 3 tiles of the clinically-validated SMN2 ISS-N1 site in its top 10 (enrichment@10 4.10 vs 0.00 for Evo 2 used as a ranker) in a pre-registered falsification test.
KALAVAI
Cooperative LLM-training protocol that fuses independently trained specialists and predicts the fusion gain before any compute is spent.
↳ +21.76% cross-lingual gain; submitted to NeurIPS 2026 with a full reproducibility audit trail.
Multi-State SaaS Platform
Enterprise park reservation and licensing system deployed across Utah, Texas, Colorado, Kansas, NJ, Alberta, and Maryland.
↳ $25M program delivered on time; zero disruption to state government operations across 7 jurisdictions.
Enterprise Cloud Migration
AWS migration (EC2, S3, Redshift) reducing costs 30% with 99.99% uptime.
↳ 30% infrastructure cost reduction; 99.99% uptime maintained post-cutover.
AI Conversational Chatbot
Tier-1 customer-interaction automation deployed at enterprise scale.
↳ 35% reduction in call volume; $250K+ in annual support cost savings.
Tech Stack
Tools I use to ship.
AI Builder & Researcher
I ship AI products.Solo.
Not prototypes. Not demos. Real products with real users, built end-to-end by one person. Every architecture decision, every deployment headache, every 3 a.m. bug fix — documented.
Build it. Break it. Measure it.
15
Projects
06
Shipped
07
Papers published
Selected Work
15 projects · 6 shipped
Research
Papers & Systems
7 published · 0 under review · 3 in progress
Live & Open Source
Show, don't tell.
Measured Outcomes
KALAVAI
+21.76%
Cross-lingual fusion gain
KALAVAI
+10.17%
Private domain fusion gain
Orion
8.5×
ANE compilation speedup
KALAVAI
99.8%
Router determinism