Enterprise SaaS · AI Design · Founding Team · 2024–presentFrom the first idea to the board review. I designed the whole journey.
Senior Product Designer on the founding team of Trmeric, an AI-native enterprise SaaS that takes demand from intake to portfolio value. I own the full design surface: IA, interaction design, design system, and production-grade prototypes that engineering builds from directly.
An AI-native B2B SaaS for demand management, resource management, and portfolio value.
Enterprises decide what to build, who builds it, and whether it delivered value, currently across emails, spreadsheets, and quarterly reviews that arrive too late to change anything.
Trmeric introduces structured demand intake, Tango (an AI agent that scopes a demand in 90 seconds instead of 3 months), and portfolio-level value tracking that closes the loop from idea to outcome.
Demand requests arrive as emails, get lost in spreadsheets, and have no lifecycle visibility. No structured intake. No AI scoping. No portfolio-level value tracking.
Intake → AI scoping → resource planning → execution → portfolio value. One continuous workflow across all personas with Tango surfacing the right context for each.
Every surface answers one question. The lifecycle is the spine; every feature is a response to a specific failure mode in how enterprises manage work. Each phase links to the work below.
Every screen is a production-grade prototype, and every tile below is real: click a shot to view it full-size, or open the live prototype it was built as.
Five deep dives. The thought process behind each one.
Demand Owner Flow
From a blank demand idea to a scoped, resourced, Tango-validated initiative.
The original onboarding dropped users onto a blank Tango screen with no context. First sessions showed users typing questions like 'what do I do here?', a blank state problem. The redesign introduced a storyboard-first approach: Tango opens with context about the user's role, surfaces the top three demand types relevant to them, and creates a pre-filled structure before the user has typed a word. The design question became: how do we make the first action feel like insight, not instruction?
Progressive disclosure (Nielsen): reveal complexity only as needed. The create-demand overlay follows a conversational arc, not a form. Mental model alignment: users don't think in fields, they think in narratives. Tango speaks narrative first.
Tango shows what it already knows about the context before asking the user to fill anything in. This shifts the first interaction from 'answer these questions' to 'is this right?', a much lower cognitive load entry point.
Every demand card carries four lines: Title, Milestone, Action, Tango input. Items that need no attention are muted gray; items requiring action are darker. The user scans the third line first. This was a deliberate attention signal system, not aesthetic choice.
Tango pre-fills the canvas with its best interpretation. The user corrects, not creates. This follows the Piaget principle of scaffolding: provide the structure, let the expert refine it. Early testing showed 81% acceptance of Tango's pre-filled scoping.
Start on the demand grid. Scan the third line of each card to find what needs action
Click any demand row to expand it in place and see the full canvas
Hit 'Ask Tango' to see the pre-flight conversation that scopes a demand in under 90 seconds
Portfolio Monitor
Multi-portfolio force graph with persona-contextual views for CIOs, Portfolio Leaders, and Resource Managers.
The Portfolio Monitor went through 23 iterations (v13 to v36). Every version that added more features was rejected. The v36 that shipped had fewer visible controls than v13, but each control carried more context. The breakthrough came from a single directive from Siddharth: 'less is more.' Not less data. Less interface. The challenge was showing the full portfolio across all lifecycle phases to three completely different personas (Portfolio Leader, CIO, Resource Manager) using one unified surface that adapts rather than fragmenting into separate dashboards.
Miller's Law (7±2 items in working memory) drove the control consolidation. Persona-contextual design: one surface, multiple lenses. The force-graph representation draws on network theory: clustering reveals relationships that a table cannot.
v13 had three rows of filters, sorts, and view toggles. Each row added one capability but increased the perceived complexity of the page. Consolidating into a single dock-style control bar with icon pills reduced visual weight by 60% while preserving all functionality.
From a June 2026 UX sync: 'You are either in dashboard mode or in action mode.' The dual-mode approach lets users toggle between viewing portfolio health and acting on it from the same surface, avoiding the context switch of navigating to a separate 'actions' page.
The portfolio header changes based on the active persona. A CIO sees portfolio-level value metrics. A Resource Manager sees capacity and utilisation. A Portfolio Leader sees initiative status. Same data, different emphasis, always sourced from the same underlying graph.
Use the category pills in the control bar to filter by lifecycle phase
Switch persona (top right). The page reorganises for CIO vs Portfolio Leader vs Resource Manager
Hover any node in the force graph to see which initiatives are connected
Project Manager + RAID
Project status and execution management built on one principle: Up-level is the only path forward.
The Project Manager is built around a single idea: the only action that matters is moving a project forward. Every element on the page either shows the current state clearly or enables the next stage advance. The RAG severity system (Red/Amber/Green) is implemented not just as color but as animation: a green project is calm, an amber project subtly pulses, a red project vibrates. This means severity is perceptible in peripheral vision without reading any text. The RAID module (Risks, Assumptions, Issues, Dependencies) lives alongside the project status as a peer surface, not an afterthought.
Exception-based management: only show what deviates from plan. Attention economics (Davenport & Beck): the scarcest resource in enterprise software is not information but attention. Animation-as-severity borrows from auditory design: a sound that gets louder is a warning, regardless of pitch.
Color alone fails for colorblind users and is invisible at a glance when scanning many projects simultaneously. The sparkle icons animate differently by severity: green glows softly, amber pulses, red vibrates. Attention is drawn physically to what needs it, before the user has consciously focused.
Earlier versions had inline editing, a status bar, and multiple update flows. Each created inconsistency in the data and confusion about what 'updating' meant. Collapsing all updates into one modal, the Up-level dialog, created a single truth point and removed ambiguity about where to go to change anything.
Items on track have light, muted text. Items needing action have darker, heavier weight. This is a deliberate reversal of typical UI defaults (where everything is equally visible). The user's eye moves to the darker elements automatically, no badge counts needed.
Notice the sparkle icon colours on each project row. Green is calm, amber pulses, red vibrates
Click 'Up-level' on any project to open the single update modal. This is the only way to change status
Open the RAID tab to see how risks, assumptions, issues, and dependencies are tracked alongside the project
Trucible · Knowledge OS
Wiki and Explorer modes for an enterprise context management system. Not a document store.
The brief from Siddharth: 'Trucible is a context management system, not a document management system.' That single sentence changed the entire design direction. A document store is a filing cabinet: you put things in and retrieve them. A context engine actively understands what matters and makes that knowledge available to Tango across every surface. The visual language had to reflect this: Trucible doesn't show folders and files, it shows relationships and understanding depth. The context strength tiers replaced a percentage score (67% complete) with a living metaphor (Seedling to Warming Up to Deep Root).
Nonaka & Takeuchi knowledge spiral (1995): tacit-to-explicit knowledge conversion. Trucible makes organizational tacit knowledge (what people know but haven't written) explicit and searchable via Tango. Information architecture: card-based progressive disclosure. The dual mode splits reading-to-understand from browsing-to-find.
A '67% complete' score creates anxiety about the missing 33% without indicating what matters. 'Warming Up' tells you the context is actively growing and usable. 'Deep Root' tells you Tango can draw on this confidently. The metaphors come from plant biology, the same language as the Trmeric brand. This was not decoration; it was alignment.
Wiki mode shows Tango-generated articles, structured, readable, sourced. Explorer mode shows the knowledge as a force-directed graph, folder tree, or list. The switch between modes is the same context, different lens. A knowledge worker reading to understand uses Wiki. A knowledge worker searching for something specific uses Explorer. Both are valid entry points.
Before a user commits to reading a full article, a preview panel surfaces the three most relevant facts from that context. This is borrowed from search engine result snippets, the preview earns the click. In enterprise context, time is the scarcest resource.
Switch between Wiki mode and Explorer mode using the toggle at the top
Notice the context strength tiers. 'Deep Root' means Tango can draw on this confidently; 'Seedling' means it's still growing
Click any source to see the connected status dot. Live sources glow green
Signals · D3 Force Graph
An Obsidian-style force-directed signal map for early-warning detection across a project portfolio.
Signals began as a question: what if you could see all the early warnings in a portfolio at once, not as a table of flags but as a landscape of risk? The D3 force-directed simulation with 79 nodes creates natural clusters that reveal structural risk patterns: a cluster of red nodes means a systemic problem, not isolated issues. The interaction model evolved significantly: the original 'click to bloom' interaction expanded nodes on click, which lost context. The new 'hover to highlight' interaction draws the neighbour graph around any node you focus on, so you always know where you are in the full map.
Network theory (Barabási, 2002): scale-free networks have hubs that, if disrupted, cause cascading failures. Signals visualises the same structure in a project portfolio: the most-connected risks are the highest-priority. The force simulation uses Fruchterman-Reingold to produce natural layouts that humans can learn to read without training.
Click-to-bloom: you click a node and its connections expand outward. Problem: you lose the position of that node in the overall map. Hover-to-highlight: the node you hover shows its neighbour graph, dimming everything else. The full map stays visible. Borrowed directly from Obsidian's graph view, the pattern already existed and users intuitively knew it.
High-severity nodes have higher 'weight' in the force simulation, so they attract more connections and cluster toward the center. This means the most critical risks are structurally visible (center-dense) before the user has read any labels. The severity color still exists, but the physics reinforces it spatially.
Signals is not a standalone tool. It lives beside the Project Manager. A signal click surfaces the relevant project card inline, so the user can act on the risk directly from the map without leaving the visualization context. The transition from map to action is zero-navigation.
Hover any node. The neighbour graph highlights around it while the rest dims
Look at where clusters form naturally. Dense clusters reveal systemic risk, not isolated issues
Click a node to surface the related project card inline, without leaving the map
Built from the turmeric plant.
Rhizome oranges. Powder golds. Leaf greens. Root browns. The palette references an actual object. That constraint made every color decision easier and the system more coherent.
One rule: restraint over decoration. No heavy left-border accents. No accent rails used as decoration. Hierarchy does the work, not ornament.
The AI gradient (#8b5cf6 → #FFA426) appears exclusively in Tango contexts. When users see it, they know intelligence is active. Mixing orange into both buttons and AI collapses this critical signal.
Pills with border-radius: 9999px. Primary = solid orange, used once per region. This keeps the UI approachable and prevents visual hardness in a dense data product.
Status, priority, and readiness are communicated via chips with tinted backgrounds. The surface stays neutral. Color earns its place only when it carries meaning.
The mark is not designed. It is grown.
18 cross-sections of real turmeric rhizomes, traced, averaged, and superimposed. Five rings sized by the golden ratio, carrying five brand promises. The full derivation, from actual plant to final lockup, has its own page.
"Prototypes are the spec. Engineering builds from them directly. That is the intended workflow, not an accident."
Aravind J · Trmeric · 2024–present
Five deep-dive case studies. Twelve live prototypes. One design system. One grown mark.
A two-sided labour marketplace that connects India's informal workers directly to employers — no contractor, no wage skimming.
← All work