श्रुति śruti · that which is heard
New York City
Judging, reviewing and honours: MIT $100K, NeurIPS, EMNLP, SciPy, Manning, Apress, BPB Publications, University of Mumbai, ABU Robocon, Forbes Editor's Choice. Across the journey: Meta, Telus, TIAA, Hiver, Flexport, Harvard, Epic Games, Hypersonix, Amazon, KFC, Taco Bell, Dapper Labs, Outcomes4Me, Eiosys, Arbor.
Ashish Dsa is Co-founder and CTO of Arbor, a Voice AI company in New York City whose product, Umi, interviews frontline workers at scale for workforce intelligence. He has spent more than a decade building software for production. Before Arbor, he held engineering roles at Meta and Telus.
His work centres on production agent harnesses for real-time Voice AI: multi-agent orchestration, tool routing, context engineering, and LLM evaluation built to hold up under enterprise load, from eval harnesses to LLM-as-judge systems.
Beyond Arbor, he is a technical reviewer of AI books for Manning, BPB and Apress Publications, has judged the MIT $100K, serves as a Program Committee member for the LIGHT workshop at NeurIPS 2026, and reviews for SciPy Proceedings, the REALM workshop at EMNLP 2026, and the LP4FM, RAAAI, RTCA, TAE, ATTRIB, AgenticOS and WiML workshops at NeurIPS 2026.
Founded in 2023 in New York City, Arbor builds Umi, a Voice AI that runs interviews with frontline workers, the people legacy surveys never truly reached, and turns what they say into workforce intelligence for the enterprises they work for.
Frontline workers speak in their own voice, participation that legacy survey instruments do not reach. In February 2026 the company announced a $6.3M seed round led by 645 Ventures, joined by Next Play Ventures, Chaac Ventures and Comma Capital.
Teams that put Umi in front of the people who do the work.
The February 2026 $6.3M announcement, led by 645 Ventures, was carried across independent industry outlets.
Ten years, seven cities, and a steady climb from intern to founder, every step of it in production engineering. The record reads newest first.
Voice AI that interviews frontline workers at scale. Technical co-founder of the company whose product is Umi.
A second founding seat, held alongside the main act.
Staff-level engineering at one of Canada's largest telecoms.
Founded and ran a software development shop.
Consulting for Dapper Labs, Swyft and Outcomes4Me.
Engineering at global scale, working remotely with the London org.
Led 23 engineers, serving end clients including Amazon, KFC and Taco Bell.
Engineered an email-sync engine moving 1M+ emails a day, on a platform used by teams at Flexport, Harvard and Epic Games.
Built VRA2, a vehicle-routing algorithm.
Mumbai.
In the old assemblies, knowledge was tested aloud before one's peers. Ashish sits on the modern ones: judge of the MIT $100K, technical reviewer for three AI book publishers, a Program Committee member for LIGHT at NeurIPS 2026, and a reviewer for SciPy 2026 Proceedings, REALM at EMNLP 2026, and LP4FM, RAAAI, RTCA, TAE, ATTRIB, AgenticOS and WiML at NeurIPS 2026.
Judge
MIT $100K
Program Committee
NeurIPS 2026
LIGHT, deployable small foundation models, Paris, December 2026.
Reviewer
LP4FM · RAAAI · RTCA · TAE
Linguistic Principles for Foundation Models (LP4FM), Resource-Aware Agentic AI (RAAAI), Real-Time Conversational Agents (RTCA), and Trust-AI-Eval (TAE) at NeurIPS 2026.
Reviewer
ATTRIB · AgenticOS · WiML
Attributing Model Behavior at Scale (ATTRIB), AgenticOS, and Women in Machine Learning (WiML) at NeurIPS 2026.
REALM at EMNLP 2026. SciPy 2026 Proceedings.
Technical reviewer
Manning · BPB · Apress
AI books, read for technical accuracy ahead of publication. He has also peer-reviewed AI research papers.
Hackathon judge
Toronto and New York City Tech Week
Innovation hackathons with leaders from Google DeepMind, xAI and NVIDIA on the panels, and with CIBC, TTC and EllisDon.
"I've competed in and won hackathons myself, and I want to give back by helping the next wave of builders get thoughtful, useful feedback on their work."
Editors publish him. Two Forbes Technology Council bylines, one of them a Forbes Editor's Choice, two expert interviews, and forty-one placements across twenty trade outlets this past year.
Published Forbes Editor's Choice for the Council essay on voice-cloning fraud, one of two permanent Forbes Technology Council bylines. Both run below, as they were published.
Featured in Inc. and Entrepreneur, profiled in TechFundingNews, and named in the Wall Street Journal and Fortune.
His own pieces, as they ran.
Two Council essays, in his own lines.
The same class of technology I use to help front-line workers do their jobs better is being weaponized against anyone with a phone.
That five minutes is the whole defense. The voice was never the real attack. The urgency, the authority, the secrecy and the irreversible payment rail were.
"I dug into the numbers" is human language. "I delved into the multifaceted tapestry of the numbers" is likely AI.
A machine generates for predictability and consensus. Humans leave their traces. The more you search for them, the clearer the machine will sound.
An eight-question solo interview by Faustine Ngila, AI Editor at Impact Newswire, published 4 August 2026, and a Founder Focus interview by Pranav Balakrishnan at Tech in Asia, published 17 August 2026.
I ship production Voice AI for frontline workers in enterprise settings. Low-resource African languages show up in demos far more often than they hold under real load.
One global average hides who the system fails.
People are more vulnerable with AI companions than with friends, sometimes even therapists. A human can judge you. With an AI, you feel you have a free pass.
The people at the top of a company often do not know what is happening on the ground. The people on the ground usually do. The question is whether companies can finally hear them.
What editors ran, in his words: verbatim passages from the published pieces, each with the idea it carries.
Algorithms, demystified Grit Daily · Decoding Social Media AlgorithmsThe mystery dies the second you've built one. It's not magic. It's just math, plus a metric somebody chose.
Postmortems over commentary BlockTelegraph · Staying Ahead of the CurveYou learn more from one well-written incident breakdown than from a month of surface-level commentary. It reminds me a lot of running distributed systems in production: the clean architecture diagram tells you how something is supposed to work, but the failure report tells you how it actually works.
Build for the worst hour Financial Tech Times · Automated Rebalancing: Client Success StoriesThe best systems are not the ones that look clever in a demo. They're the ones that behave sensibly when the operator is distracted, emotional, or overloaded.
Founder time, re-architected Grit Daily · Time Management Tips for Growing BusinessesLatency had to stay under 800 milliseconds or the AI stopped feeling human, and I was personally burning 30 hours a week debugging packet rates and jitter with the engineers. I was running my own time like a single-threaded process while the business fired concurrent requests at me.
Automate friction, keep empathy human College Recruiter · 18 AI-powered hiring systems candidates preferThe tradeoff worth being honest about: candidates still rate the conversation as slightly less "natural," and most people still want a human at the emotional moments, like a real rejection with feedback. The systems candidates genuinely like are the ones that automate the friction but keep a human for the parts that need empathy.
Small teams, hard deadlines CEO Official Magazine · Staying AgileNow when something critical comes up, we find the technical issue, give it to a small team with a 48-hour deadline, and let them work. Small teams shipping code beat a big group trying to plan everything.
Speed is the startup's moat Freeduhm · 11 Misconceptions About Billion-Dollar StartupsYour only real advantage over a giant is not capital, it is how fast you can execute a full loop, ship, test, learn, before they have finished a planning meeting. Big players drag every change through policy and approval layers.
Boundaries follow ownership CTO Sync · Choosing the First Service BoundariesThe first seam I look for is not the most "technical" boundary. It's the one with the cleanest business ownership and the lowest coordination cost. If a service still needs three teams to change it safely, it's not really a service yet, it's just a distributed monolith with extra networking problems.
Cut what the customer never sees COO Insider · Protect Customer Experience During Cost CutsIt's common for leaders to try to cut 10% of costs from every system, but this just makes the entire operation 10% worse. To protect the customer experience, you need to focus on the exact interaction point the user feels, and aggressively cut the cost of the background work they never see.
Standardize the contract, not the stack CIO Grid · Set Smart Guardrails in Enterprise ITWe adopted a formal containerization contract. "Ship a Docker image, make it health checkable, and integrate into our Kubernetes workflow." That was it.
AI drafts, humans finish PRAPI Research · How founders actually run their companies on AII run an AI company, so I'm a heavier user than most, but the honest list is shorter than you'd think. What stuck: a coding agent for the first pass, an LLM for drafting and triage where a human does the final pass, and AI transcription on every meeting. What I dropped: most standalone AI-for-X point tools, they get eaten the moment the model we already pay for can do the same thing in-house.
Editors have run his commentary in the technology trade press more than forty times this past year, across twenty outlets. Open an outlet to read the published pieces.
B.E. in Computer Science, Fr. Conceicao Rodrigues Institute of Technology, 2017. Awarded the gold medal for his final-year project.
The yantra is the instrument built with precision so the work can flow through it. These are his.
Voice AI and real-time speech systems; voice model training, fine-tuning and evaluation; NLP and conversational AI at enterprise scale.
Multi-agent systems, agent orchestration and agent harnesses; tool-using agents and context engineering.
LLM evaluation and eval-harness engineering; LLM-as-judge systems; real-time inference pipelines.
Model deployment and MLOps; retrieval-augmented generation; big data and analytics processing millions of data points.
Voice-cloning and deepfake fraud defence; SOC 2-compliant enterprise platforms serving Fortune 500 customers, including Comcast.
Python: PyTorch, scikit-learn, FastAPI. Cloud on AWS and GCP. Full-stack with React and Next.js.
Born in India, educated in Mumbai, and shaped by a coastline of languages. Ashish grew up speaking Konkani, his native tongue, and carries six more: English, Hindi, Marathi, Punjabi, Gujarati and Kannada.
It is no accident that a polyglot from Mumbai ended up building machines that listen. When you grow up switching between seven languages, you learn early that being heard, properly heard, is not a small thing.
The page you are reading is set like a pothi leaf: geru red oxide and turmeric gold on handmade paper, with toran, kolam, Warli and jaali as the grammar of the frame. The instruments are modern. The listening is ancient.
For speaking, judging, reviewing or press:
645 Ventures
Next Play Ventures
Chaac Ventures
Comma Capital
