Today on the Salesforce Admins Podcast, we talk to Parth Sevak, Director of Technology and Principal Architect at Incepta. Join us as we chat about how Parth built a multi-agent system designed to connect surplus food with the people and organizations that need it the most and won the Agentforce for Good Grand Prize at the TDX Hackathon.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Parth Sevak.
If you listened to last week’s episode with Alexandra Iyer, you know that Agentforce for Good was a popular problem at this year’s Agentforce Hackathon at TDX. Contestants took on big issues like nonprofit volunteer coordination and disaster relief. That’s why I was so excited to sit down with Parth Sevak, whose project Harvest Bridge won the Agentforce for Good Grand Prize. Harvest Bridge is a multi-agent application that connects food donors with organizations near them.
As Parth explains, food waste is a serious problem. According to the UN’s World Food Programme, about 318 million people are facing acute hunger today. “In North America, 30-40% of the food that is produced never gets eaten,” he says. So he decided that this would be the perfect problem to tackle for the Agentforce Hackathon at TDX.
Under the hood, Harvest Bridge features multi-agent coordination between four agents to handle donor intake, food matching, volunteer logistics, and reporting analytics. While it sounds incredibly complicated, Parth is quick to point out that 80% of the work was done in configuration with out-of-the-box admin tools.
Parth needed to write some Apex to do specific things like geo-matching, which he vibe-coded with the Claude plugin for Agentforce. Crucially, he didn’t have to write glue code to make everything work between Agentforce, Data 360, automations he built in Salesforce, Slack, and Tableau. “All of it just worked like a charm,” he says, “five years ago, that integration story would have been months, if not years.”
In just a few days, Parth was able to build an autonomous, multi-agent system that uses Agentforce, Data 360, Slack, and Tableau to match surplus food with local organizations and coordinate delivery in under 90 minutes.
If you’re looking to get started with Agentforce, Parth recommends jumping on Trailhead as your first step. The Agentforce Specialist certification gives you the tools you need to start building, and then it’s all about getting your hands dirty.
Make sure to listen to the full conversation with Parth Sevak about how he built Harvest Bridge and won the Agentforce for Good Grand Prize. And don’t forget to subscribe to the Salesforce Admins Podcast so you never miss an episode.
Mike:
This week on the Salesforce Admins Podcast, we’re talking with Parth Sevak about how a real world food waste solution went from idea to working system in just days using Agentforce. Drawing from a challenge that impacts millions globally, Parth built HarvestBridge, which is a multi-agent system designed to connect surplus food with the people and organizations that need it most. Now, this was part of the TDX26 Hackathon Challenge and what makes this conversation the most compelling isn’t just the technology, it’s how admins and architects can really orchestrate data automation, AI agents and human coordination together without months of integration work.
We also get into what it means to design a trustworthy system where humans stay in control while AI handles scale and speed. So, if you’ve ever been wondering about how Agentforce changes the role of the Salesforce admin from builder to system orchestrator, this episode’s for you. Be sure to subscribe, share the episode with your team, your friends, your local Salesforce admin user group and let us know what kind of real world problems you’d solve with Agentforce. But for now, we’re going to get Parth on the podcast. So, Parth, welcome to the podcast.
Parth Sevak:
Thanks, Mike. Really glad to be here.
Mike:
Well, I’m glad to have you. So, if everything shakes out, the episode before this will be the episode with the Agentforce for Good People, but sometimes scheduling is what it is. But I was at TDX this year and we had Hackathon winners and you were part of the Agentforce Hackathon. And so, that’s how I got connected with you. But I think before we get into that, I’d love to know a little bit about how you got started in the Salesforce ecosystem, what you do and let’s go from there.
Parth Sevak:
Absolutely, Mike. Again, thanks for having me. So, the story of how I ended up in Salesforce is really interesting. That was not my plan. In fact, back in 2011, I was a fresh computer science engineering grade hunting for my first real tech off somewhere. And that time, the startup economy was just taking off and Java was the thing, the only thing in my perspective as far as I was concerned. And then my boss pulled me aside and said, “Parth, you are going to work on Salesforce.” And I said, “Oh, what is Salesforce? I’m looking for Java opportunities.” So, he made his pitch. “Apex is basically Java, you will feel right at home. And there is this thing called Dreamforce, massive event, you might get to experience it someday.”
And that’s where you see a lot of enterprise innovation is going to come in. Basically, he sold it and 16 years later, here I am loving the challenge, loving the stretch, always finding ages I haven’t touched yet and it keeps me motivated.
Mike:
He was a heck of a salesman. Java, here’s Salesforce. It’s just like Java. And if you’re good enough, you might get to go to Dreamforce. That’s awesome. So, go ahead.
Parth Sevak:
Yeah. And just to tell you what currently I’m doing. So, these days-
Mike:
Yeah, please.
Parth Sevak:
… I’m working as a director of technology at Incepta Solutions where I lead data, CRM integration and agentic AI transformation for enterprise clients across financial services, retail manufacturing and pharma healthcare. But I make sure to keep one foot in hands-on building always. And that’s how HarvestBridge happened at TDX, as you see.
Mike:
Yeah. So, let’s talk about that. I mean, it’s really cool. This isn’t our first hackathon since we’ve come out with Agentforce. We’ve been doing hackathons and the early hackathons, I was a judge in quite a few and some of the really neat ideas that people were coming up with for the use of AI agents and some agentic use cases I think were really kind of interesting. I know we had that at TDX. So, I guess let’s start off with the TDX Agentforce for Good Hackathon. What made you want to enter that to begin with?
Parth Sevak:
Yeah, to be really honest, it was again, my boss encouraged me because even I was not sure whether I would be attending the TDX in first place, but my boss was really motivated to join the party and he invited me. And then I realized, okay, if I register for TDX, there is an option to participate in the hackathon. And that’s how my journey began.
Mike:
Oh, well, we made that awfully easy for you, didn’t we?
Parth Sevak:
Yeah.
Mike:
I mean, I would like to think I’m a competitive person, but when it comes down to it, I think I really like watching competitions as opposed to competing. What made you think, “Oh, I’m going to get into this because the idea that I have is so great.” So, tell me a little bit about what your idea was and what it solved for.
Parth Sevak:
Yeah. So, honestly, the topic I picked is not a fleshy topic, right? I picked the food waste, which is definitely not a hackathon topic. It’s not a biotech, it’s not a blockchain. But my heart said, pick it anyway because food is the most basic thing and somehow we managed to throw away nearly half of it while families a few blocks away are skipping dinner. And then again, I am a data guy, so please allow me to share some numbers here.
Mike:
Oh yeah, please.
Parth Sevak:
And that application for me, that 2.5 billion tons of food wasted every year, 783 million people are facing hunger. And when I read that and thought, “This cannot be right.” And I check UN website, food banks’ websites, and again, the numbers held up. And the on that hit really hardest is this number, which is in North America, 40% of food produced never gets eaten right here in our cities, whether it’s San Francisco, Toronto, basically with our neighborhood.
Mike:
I mean, that’s crazy. I know I’ve seen … I forget the show. It was on Discovery Channel here in the US and they talked about … I didn’t know there were hog farmers out in Las Vegas, but there are and they work with the casinos. The casinos have all those buffets to get all their food scraps. I thought that was really, really fascinating. I guess you think about it and until you look at it on a larger scale, do you even understand how much food is produced and unfortunately how much is wasted?
Parth Sevak:
Yeah, totally, Mike.
Mike:
So, let’s talk about your solution because it involves Agentforce. So, were you going to feed Agentforce some apples? That’s a joke because you can’t feed it apples. I mean, you could feed it apples maybe like an emoji.
Parth Sevak:
So, yeah, I think let me first of all piggyback on this one, this idea that this HarvestBridge as a idea, right? When you look at this project from the outside, it may sound like as a big AI project. We have four agents, multi-agent coordination, the whole team, but the reality on the inside is three of those four agents are about 80% configuration. And this is the Agentforce play in my opinion because I used agent builder topics, instructions, guardrails, then flow for orchestration, custom objects, validation rules, formula fields, like typical standard admin toolkit. And obviously, there was a 20% which is code, which is attacks actions for things like geo-matching.
So, that also like I wipe coded with cloud plugin for Agentforce available. And what came back was the quality of a season developer with proper design patents, clean error handling, governance limit aware as I made sure to provide the right context. That is very important. So, anytime if anybody ever built a flow with a decision branch and a few validation rules, you can think the way you need to build an agent. The tools just caught up with you.
Mike:
Yeah. Well, actually we had a few presentations at TDX and the admin track from admins who are vibe coding and I’ve had a few people on the podcast talking about how they’ve vibe coded and just the product itself is just insane to think you can sit down and there’s a plan and there’s an action button. And even in the plan button, I was really impressed with how much Agentforce Vibes could give you back as an admin to really help you plan your application.
Parth Sevak:
Yeah, that is really empowering thing has happening just in front of eyes and not just that, but we are at the forefront of that innovation and definitely the Agentforce for Good is really the category that should motivate everybody to get their hand dirty for something to produce something really good.
Mike:
Yeah. So, let’s talk about while you were building your application to deal with food waste, what was a point in time when you sat back and maybe learned something that you didn’t know as you were building it?
Parth Sevak:
Yeah. So, before I answer that question, I need to tell you how I approach the build process if it is okay.
Mike:
Oh, please.
Parth Sevak:
Yeah. Okay. So, basically like Mike, to be very frank, I didn’t follow a plan. I would love to say that I sat down on day one and map out a clean sprint like user stories and backlog, but that was not what happened. What actually happened is I just started building right away. So, I picked the first thing, which is a donor bot. This is the agent that talks to donors and captures surplus food. First I got that working and the moment it ran, the next question was obvious that, okay, now surplus is in the system in the Salesforce. Now where does it go? So, I built the matchmaker agent and once matchmaker was written in matches, the next question came up on its own that who actually picks the food up and delivers it?
Who is the volunteer? That became logistics coordinator on Slack and by then, I had data flowing through the whole system and it was almost asking to be visualized and that become impact analyst on Tableau. So, overall, I ended up building four agents, each one revealed the next. And what made that look possible beyond Agentforce itself is that I had AI assistance alongside me the whole way, like not just for the apex spaces we talked about, but for the thinking too, like should this be one agent or two? What edge cases am I missing? Is this topic structure clean, less a solo build and a more long conversation with a partner who is read every Salesforce document kind of like ever published.
So, when I started building on the Agentforce platform, I had some high expectations and let me tell you, like what really surprised me in a pleasant way is the breadth. I use Agent Builder for the agents themselves, the topics, instructions, custom actions. Then underneath, I relied on agent script for the deterministic safety rules. So, things like allergens and expiry windows are not left to the LLM to decide. From there, I brought an Agentforce voice for the call-in donation capture, which I think just became GA recently, which I think, yeah, an intelligent context handled passing food details out of photos that donor send in.
All of it was sitting on Data 360, which held the unified profiles of donors, recipients, and volunteers. Here is the part that genuinely caught me off guard. I didn’t write any glue code between those layers, like an agent would hand off to a flow which will call an apex action that will update the records in the Data 360 and that would trigger a Slack message. That would feed back into another agent’s context and all of it just worked like a charm. And if I reflect like five years ago, that integration story would have been months, if not years and this made me possible to deliver something so amazing in just few days and that was my wow moment from the whole build.
And Mike, if I just like tell you another perspective that the other thing I was nervous about going in was hallucination because I am building four different agents and then you have obviously hear the horror stories, right? But between agent builders topic structure, agent script for hard rules and clear instructions, that’s how I make sure that my agent behave very well. So, food safety rules around allergens and expiry are not something that I would let an LLM to decide.
I would make sure to enforce them deterministically at the architecture level and for that I leverage agent script and thank God, that agent script work really well and where I would love to see things improve is just the debugging because when an agent’s reasoning across topics and data sources, tracing what it actually did is still a bit difficult. So, the tooling is getting better, but its place still I’m watching.
Mike:
Yeah. I do an Agentforce Now workshop every month and I always try to point out when we’re building it and preview how you can go through and see what the agent is doing and how it’s deploying its sub-agents. And I think with everything we’re still learning what AI is and how to debug it and it’s this interesting intersection because we want the intelligence and the context and the engagement of AI, but I think people got really hooked on how perfect a bot would respond and a bot would respond because it’s always A, B or C and we gave it those answers.
And the nice thing about AI is it doesn’t necessarily have those answers, but it can think independently, but we still want that control. So, it’s like building the perfect employee.
Parth Sevak:
Yeah, absolutely. And that’s why, like as I said, Mike, the deterministic approach is very significant and particularly, I would kind of rely on agent script that really gives that control back to the human or admins or architects so that they don’t need to basically do the guesswork that, okay, what will be the response of agent if I do this prompt versus that prompt? It can simply be predictable. So, that is the cool thing I would say about the agent script and what it makes possible now in the reality of governance.
Mike:
So, I think one thing that’s really cool about hackathons is it’s kind of blue sky and green field. You can come in with a solution and like in your case, it really solves a real world problem. If somebody is listening to this and thinking, “Wow, I’d really like to get started with Agentforce,” what was the first thing that you did to get hands-on with Agentforce and start understanding the product to the level that you do now?
Parth Sevak:
Oh, I think I don’t recall exactly the date and time, but I think it was I think a couple of years ago when there was already like Agentforce discussion has already began because I was active in the community and thanks to Trailhead. And I basically started my journey to equip myself about like what is this best Agentforce, right? And then I did few modules and ultimately I think last December, I got certified in the Agentforce as Agentforce specialist.
So, that really gives me clarity that okay, just not about like, okay, what I can build for the demonstration purpose, but also for me, it’s very important that I can actually pass on that knowledge to my team so that they can ship the products or ship the agents that is really the kind of like configurable and scalable across the systems, not just one particular cloud or system or CRM.
Mike:
Yeah. I mean, anytime I need to learn something for a demo or for a presentation I’m putting together, I usually turn to Trailhead first and kind of walk through it and get an idea of what the product can and do and how I can configure it. So, I’m right there with you. Even Salesforce people still learn on Trailhead.
Parth Sevak:
Yeah, that’s amazing actually. I call it as a live report, it’s always live. You never get any stale recipe or stale dishes. So, that means everybody, whether it’s a employee of Salesforce, partners or customers, everybody can rely on Trailhead for the real time information for sure.
Mike:
So, moving forward, we have a little event in the fall that we do called Dreamforce that you mentioned. Obviously, none of us have a crystal ball. Well, I’m sure a few people in Salesforce do to know what we’re going to show, but what are some of the things you would love to see this year at Dreamforce?
Parth Sevak:
Okay. So, as I said, I have been part of the Salesforce ecosystem for almost 16 years and what really amazing to see is that the platform never sits still. So, when I started, here is the context, Mike, what I’m talking about is when I started we were building the S controls and it came to Visualforce.
Mike:
Yeah, that’s back when I started too, called them Superman controls.
Parth Sevak:
Yeah. Then Visualforce, like proper MVC, controllers, pages, then Aura framework. And my first test of real component-based DUI was that Aura and then Lightning Web components. And I remember thinking, okay, this changes everything. And now 2026, we have got React running on Lightning Web Runtime along with the Headless 360 and agents that reason and act on their own and every leap has bigger than the last, right? So, it is the one ecosystem I have worked in that’s never stopped becoming more interesting.
I will make sure like a kid, I always go without any expectation that, okay, oh, what I’m going to get out of this conference because I know whatever will be there, it will be super amazing, super productive and make everyone’s life much better than it was before.
Mike:
Yeah. Wow, that sets the wheels in motion to draw the bar even higher for all the rest of us. I appreciate you sharing that. And I will say one thing that we do at Dreamforce every year is we not only have volunteering activities, but I know the entire event is very conscious about the food that we use and how it gets recycled and dispersed evenly to make sure that we have the right amount. Because conferences I’m sure are big places where a lot of stuff can go to waste as well.
Parth Sevak:
Yeah, no absolutely. And I couldn’t be more prouder than this, particularly like I was not also that conscious about all these basic problems, but then I started volunteering. I involved with some foundations and then I realized, “Oh man, wow, we are living in a 21st century and still we are not able to figure out the solution for this day-to-day problems.” And it really kind of like hurts in our heart like if we see. So, basically like my belly is full. It’s not solving the problem, right? I cannot be that selfish and truly like I couldn’t agree more like Dreamforce. And that’s I think really set the high bar that how the conferences should be run from the humanitarian angle as well.
Mike:
Yeah. Well, thanks for the kind words, Parth. I appreciate you coming on the podcast. This was a great conversation. I will say this now I should be able to find your TDX video and post that in the link so that everybody can see the demo of the app that you built and see how cool it was because I know we like to showcase those winners and so hopefully it inspires other people to not only participate in the hackathon but bring those solutions back to their communities as well.
Parth Sevak:
Yeah. And Mike, if you permit, can I say something like I would take maybe a one minute-
Mike:
Sure.
Parth Sevak:
… which is very close to my heart and this is like a period like everybody is going through the transition. So, what I say is like, yeah, we need to be a jack of all, but master of one. And what I mean by that is that we don’t want to restrict ourselves doing the same thing for too long. It’s easy to get comfortable. For example, you are good at your role, the work is familiar, but comfort is also the thing that quietly stop you from growing. And right now with where the technology is, the agentic AI is here, right?
This is exactly the moment to step outside that comfort zone, not because that you have to, but because the upside is so much bigger than the downside and Agent AI use well is probably the most empowering tool any of us have had access in our professional journey at least. So, I can take you somewhere you couldn’t get along, but only if you handle it right, it can. So, stay the architect, keep judgment in your hands, let it stretch you instead of replacing you. The individuals who lean into that, who do the thing that is needed most for their clients, their employers, and honestly for their own sake, those are the people I think come out of the next two or three years transformed.
So, for me, what is next is more building, more sharing, more conversations like this one. And if one person listening today challenges themselves to try something they have been putting off, that is a great month.
Mike:
I would agree. Couldn’t have said it any better than you did. Well, Parth, thanks for coming on the podcast. It was great having this conversation with you.
Parth Sevak:
Yeah, thank you, Mike. This was a real pleasure.
Mike:
Huge thanks to Parth for joining us and sharing how HarvestBridge came together at the intersection of AI, automation, governance, and real world impact. I know one of the biggest takeaways from this conversation is that the future admin isn’t just configuring tools, they’re designing systems that connect people, data, agents, and outcomes responsibly. So, whether you’re experimenting with Agentforce for the first time or already building multi-agent workflows, the opportunity is bigger than just productivity. It’s about solving meaningful problems faster and with more intention.
Of course, if you enjoyed this episode, subscribe, leave a review, share it with somebody who’s ready to start building what’s next. And until next time, we’ll see you in the cloud.
Today on the Salesforce Admins Podcast, we talk to Joshua Birk, Senior Director of Admin Evangelism at Salesforce. Join us as we chat about how Agent Script helps admins build more predictable and reliable AI solutions. You should subscribe for the full episode, but here are a few takeaways from our conversation with Joshua Birk. […]
Today on the Salesforce Admins Podcast, we talk to Jennifer Lee, Lead Admin Evangelist, and Eliza Riley, Content Manager for Admin Marketing at Salesforce. Join us as we chat about what’s coming at TDX and how to get the most out of your visit. You should subscribe for the full episode, but here are a […]
Today on the Salesforce Admins Podcast, we talk to Melissa Hill Dees, nonprofit Salesforce consultant and Salesforce MVP. Join us as we chat about how her TDX Hackathon team built a conference scheduling agent from scratch in 16 hours. You should subscribe for the full episode, but here are a few takeaways from our conversation […]
