Live · Axestrack control tower · 1,000+ ai calls/day
go, pal! · a small studio · est. 2023
Active · Mashreq Bank · PEP + Marketing agents shipped
Vishvakarma Bhawan · IIT Delhi
Dubai Government · board-meeting agent deployed
4 founders · 12 engagements shipped
Now booking q3 — 2 client slots open
Momentum over noise.
A small studio · est. 2023
Vishvakarma Bhawan · IIT Delhi

AI systems.
Marketing data.
Built deliberately

A four-person studio. We ship the systems that move decisions, then leave.

Est. 2023 — Vishvakarma Bhawan, IIT Delhi
12 engagements shipped
Average team size: 4
Avg. engagement: 8 weeks
01
Voice AIEdTechFine-Tuned Persona
NV.AI — Hyper-Real Hindi Educator Avatar
Motion Education (go, pal!) · Jun – Aug 2023
ReadClose
Idea
Could a student talk to their favourite teacher anytime — even when that teacher has lakhs of followers and zero spare time?
Company
Motion Education, run by Nitin Vijay (NV Sir), had a massive student base but a finite teacher. They needed an AI that felt genuinely like him — not a chatbot, not a FAQ bot. Built pre-LLM maturity, when Hindi voice AI was widely considered infeasible.
Architecture/Persona-Fidelity Engine · Fine-Tuned + Psychometric
INPUT PROFILE MODEL GUARD VOICE OUT 500h+ Video Corpus Transcripts · SOPs Teaching Annotations PSYCHOMETRIC PROFILE Vocab · Pace · Empathy Thought structure mapped FINE-TUNING · RLHF Hindi LLM Tone · Idiom · Warmth NV-Sir style aligned IDENTITY GATE Domain routing Off-topic → rejected HINDI TTS · ELEVENLABS Emotion-calibrated Urgency · Softness NV Sir supporting layer core innovation delivery surface
01Approach
Suggested Solve
  • Psychometric persona modelling from NV Sir's videos, transcripts, and teaching patterns to capture tone, vocabulary, and emotional warmth
  • Custom Hindi TTS pipeline with emotion calibration — softness, urgency, and encouragement layered onto output
  • Identity-gated response system and domain-constrained LLM routing to keep answers on-subject and in-character
02Build
Product Outline
  • Query classifier routing academic vs motivational vs personal queries to distinct response modes
  • Personality-mimicking engine: word choice, comprehension level, thought structure modelled on NV Sir's style
  • Speech synthesis tuned to his voice — the system simulated his natural conversational pace and warmth
03Outcome
Delivery & Impact
  • Lakhs of queries in the first week — zero paid marketing
  • User feedback: indistinguishable from NV Sir
  • First Hindi educator AI avatar of its kind in India, built when others said it couldn't be done
StackFine-TuningRLHFElevenLabs TTSHindi NLPPsychometric ProfilingOpenAI APIFlaskMongoDB
L+
Queries week one
0
Paid marketing
1st
Hindi educator AI in India
02
Symbolic AIMulti-Layer BrainRAG + SQL + Rules
AstroAI — 25 Astrologers Replaced by One Engine
AppsForBharat / ShriMandir App (go, pal!) · Nov 2024 – Feb 2025
ReadClose
Idea
Vedic astrology has crores of unique combinations. What if a hybrid engine could decode them computationally — more accurately than most human astrologers?
Company
AppsForBharat's ShriMandir app had 25 human astrologers handling per-minute chat consultations. Users were frustrated by slow responses and long pauses for analysis. The company needed precision at scale — and a pricing model that didn't penalise the user for thinking time.
Architecture — Symbolic-Neural Hybrid · Layer-by-Layer Routing Brain
USER QUERY Birth chart + question LAYER 1 QUERY CLASSIFIER Workflow selector SQL layer routing PLANET-HOUSE ENGINE 100s of table cross-joins Crores of unique PnC cases DASHA TIME-SERIES SQL + RAG hybrid Vedic + KP cross-relation PLANET MOVEMENT Transit calculator Ephemeris data + SQL LAYER 3 WEIGHTED SCORER Output weightage algo Cross-system reconcile LAYER 4 RAG GROUNDING Classical Vedic texts No hallucination possible LLM SURFACE Language only · 5% of system ◆ SYMBOLIC-NEURAL HYBRID — RAG + SQL + Rules + Cross-table relations. LLM is the last layer, not the brain.
◆ Symbolic-Neural Hybrid
◆ Layer-by-Layer Routing Brain
01Approach
Suggested Solve
  • Build a deterministic rule-logic layer for planet-house combinatorics, dasha trackers, and weighted scoring — going deeper than any RAG-only competitor can
  • Layer controlled LLM interpretation on top of rule outputs — generative but grounded, never hallucinating astrological facts
  • Hybrid Vedic + KP system via SQL-based layer selection handling crores of unique birth chart cases
  • Shift from per-minute billing to activity / response-based pricing — reducing user friction
02Build
Product Outline
  • Workflow selector: deep query, time-prediction, event-probability modes — routed by query type
  • Planet-house-sign combination engine with dasha tracker, planet movement calculator, and weightage algorithms
  • Response-generating agents calibrated to emotional tone and user's query intent
  • RAG retrieval grounding outputs in classical Vedic texts — no hallucinated interpretations
03Outcome
Delivery & Impact
  • 25 human astrologers fully replaced — higher user retention post-launch
  • Handles crores of unique PnC cases no simple RAG model can replicate
  • Accurate enough that users call results 'amazingly true' — part of gopal's ₹45L+ Year 1 revenue
StackSymbolic Rule EngineSQL · 100s of TablesRAG Hybrid RetrievalLangChainPythonVedic + KP LogicOpenAI API
25
Astrologers replaced
Cr+
Unique cases handled
4
Reasoning layers — LLM is Layer 4
03
Agentic ResearchFintech ComplianceMulti-Source Synthesis
PEP Compliance at Mashreq Bank — 1 Day → 5 Minutes
UnifyApps (Mashreq Bank, Dubai) · Sep – Dec 2025
ReadClose
Idea
What if PEP compliance checks — a full analyst-day of manual work — could be collapsed into 5 minutes, with higher thoroughness?
Company
Mashreq Bank (one of UAE's oldest financial institutions) had analysts spending a full working day per Politically Exposed Person verification — querying multiple databases, cross-referencing news, manually building risk profiles. Expensive, slow, and audit-fragile.
Architecture — Confidence-Weighted Synthesis · Exceptional Research Depth
Sanctions Databases Global News Archives Corporate Registries Court Records · Filings Social Signals · Web CONFIDENCE-WEIGHTED SYNTHESIS ENGINE Entity resolution Cross-source reconcile Risk categorisation CONFIDENCE SCORE Low Med High AUDIT-READY REPORT Source citations · Flags AI ANALYST COPILOT NL investigation queries 5 min was 1 full working day ◆ CONFIDENCE-WEIGHTED SYNTHESIS — 5 parallel sources · entity resolution · risk scoring
◆ Confidence-Weighted Synthesis
01Approach
Suggested Solve
  • Multi-source autonomous research engine querying databases, news archives, and public records in parallel
  • Confidence scoring and risk categorisation delivering structured findings — not raw data dumps for analysts to sift
  • Embedded AI copilot for analyst-led deep investigation — human stays in the loop, AI does the legwork
02Build
Product Outline
  • Agentic pipeline: person profiling → multi-source fetch → entity resolution → risk scoring → audit-ready report
  • Structured output with confidence bands, risk flags, and source citations — not prose summaries
  • Natural language copilot interface for follow-up questions on generated profiles
03Outcome
Delivery & Impact
  • 1 day → 5 minutes per PEP check — ~99% time reduction
  • Structured evidence with confidence scoring — production handoff to Mashreq compliance team
  • Agentic workflow end-to-end — not a GPT wrapper
StackLangChain AgentsCrewAITavily SearchOpenAI APIn8nEntity ResolutionConfidence Scoring
99%
Time reduction
5
Parallel research sources
Live at Mashreq Bank
04
Agentic AIMarketing IntelligenceFull-Team Automation
3× Marketing Output — Zero Extra Headcount
UnifyApps (Mashreq Bank) · Sep – Dec 2025
ReadClose
Idea
A fully autonomous marketing pipeline — from keyword to publish-ready asset — with no human handoff between steps.
Company
Mashreq Bank's marketing team was bottlenecked at content production. Every piece involved multiple handoffs across briefing, research, writing, brand review, and scheduling. The need was a fully integrated agentic system, not another writing assistant.
Architecture — Autonomous Marketing Pipeline · Marketing Knowledge Embedded
INPUT Campaign goal Audience · Channel KEYWORD RESEARCH Trend mining Competitor gap SEO intent cluster DEEP RESEARCH AGENT Web · News · Reports Contextual · Current Not templated CONTENT DRAFT AGENT Marketing knowledge embedded in model Tone · Format · Hook BRAND ALIGNMENT CHECKER Brand SOP baked in Auto-approve / reject No manual review CALENDAR + PUBLISH AGENT Platform formatting Schedule + launch No human handoff output ◆ 6-STAGE AUTONOMOUS PIPELINE — Marketing knowledge embedded · Zero human handoff between stages
◆ Autonomous Marketing Pipeline
01Approach
Suggested Solve
  • Autonomous pipeline: keyword generation → deep research → brand-aligned content draft → social calendar → publish-ready asset
  • Deep-research agent layer for contextual, current content — not templated outputs
  • Brand SOP encoded directly into a classifier-checker — auto-approve / reject without manual review
02Build
Product Outline
  • Keyword + intent agent: trend mining, competitor gap analysis, SEO intent clustering
  • Deep research agent: live web + news + reports, contextual and current — not boilerplate
  • Brand-alignment classifier baked into the loop — content cannot ship that fails it
  • Calendar + publish agent: platform-specific formatting and scheduling, end of pipeline
03Outcome
Delivery & Impact
  • ~3× marketing team productivity — same headcount, triple output
  • Zero human handoff between any stage of the pipeline
  • Production at Mashreq Bank — a regulated financial institution
StackLangChainCrewAITavily Deep ResearchOpenAI APIn8nBrand ClassifierCMS Integration
Marketing productivity
6
Autonomous stages, zero handoffs
Live at Mashreq Bank
05
Meeting IntelligenceAgentic AIGovTech
Board Meeting Agent — Dubai Government’s Institutional Memory
UnifyApps (Dubai Government) · 2025
ReadClose
Idea
An AI that sits in every board meeting, retains all context, flags when decisions contradict prior commitments, and actively participates when a relevant point is overlooked.
Company
Dubai Government needed an agent that wouldn't just record minutes but participate — retaining context across every prior board meeting, flagging contradictions with earlier decisions, surfacing forgotten commitments live. The bar: secure, government-grade, sensitive context, no leakage. Like a founder's office in the room.
Architecture — Persistent Meeting Intelligence Agent
LIVE MEETING Audio · Transcript Real-time stream INGESTION + NER EXTRACTION Decisions · Action items People · Deadlines · Topics INSTITUTIONAL MEMORY STORE All prior meetings indexed Commitments · Context Vector + Graph DB CONTRADICTION DETECTOR Flags decisions vs prior commitments ACTIVE PARTICIPATION ENGINE Raises forgotten points in real-time SECURE MINUTES + ACTION TRACKER Persistent · Encrypted Next Meeting ready ◆ PERSISTENT MEETING INTELLIGENCE — Context persists across every meeting, forever
◆ Persistent Meeting Intelligence
01Approach
Suggested Solve
  • Real-time meeting ingestion with NER extraction — decisions, action items, people, deadlines, topics
  • Persistent institutional memory across all prior meetings using vector + graph indexing — recall any commitment instantly
  • Contradiction detector that flags when a current decision contradicts a prior board commitment
  • Active participation engine — raises forgotten points during the live meeting when relevant and overlooked
02Build
Product Outline
  • Audio + transcript stream → NER pipeline → structured commitment records
  • Vector + graph store covering every past meeting — queryable in natural language
  • Contradiction + missed-item detectors running live, surfacing alerts mid-meeting
  • Secure minutes + action tracker with encrypted persistence between sessions
03Outcome
Delivery & Impact
  • Deployed for Dubai Government — replaces a full executive secretary and institutional records function
  • Adani Group expressed interest in onboarding the same system
  • Highest-trust AI deployment — secure, encrypted, government-grade context
StackReal-time TranscriptionNER ExtractionVector DBGraph DBLangChainContradiction DetectionEncrypted Storage
Meeting context retained across sessions
Gov
Dubai Government deployment
Adani
Expressed interest in onboarding
06
Agentic AIHumanoid VoiceLogistics Intelligence
Axestrack — 1,000 AI Calls / Day, Intelligent Control Tower
Axestrack · Jan 2026 – Present
ReadClose
Idea
Replace entire manual control tower teams with a highly intelligent parallel agent system — one that detects, decides, acts, and communicates simultaneously across the full logistics stack.
Company
Axestrack runs control-tower operations for large logistics fleets. The bottleneck was human: every anomaly, delay, or upsell call had to route through ops managers and call agents. They needed a system that could detect, decide, communicate, and execute — all in parallel — across thousands of events a day.
Architecture — Parallel Agent Topology · Highly Intelligent Control Tower
EVENT TRIGGER Delay · Anomaly From structured logistics DB ORCHESTRATOR Decision Intelligence Context · Priority · Routing Replaces ops manager HUMANOID VOICE CALL AGENT Driver · Fleet · Consigner Context-aware · Adaptive ANALYTICS COPILOT NL dashboard query SQL + LLM · Live data EXECUTION AGENT Action recs + loops Automated · Self-closing CRM + FEEDBACK Auto-logged outcomes Continuous improvement Self-learning loop 1K+ calls / day Revenue↑ Auto upsell via AI MODULAR AI BACKBONE Future full agentic execution ◆ PARALLEL AGENT TOPOLOGY — Simultaneous detection, communication, analytics, and execution
◆ Parallel Agent Topology
01Approach
Suggested Solve
  • Orchestrator with decision intelligence — context, priority, and routing handled by the agent layer, not by a manager
  • Humanoid voice call system for drivers, fleet owners, consigners, ops teams — adaptive and context-aware, not scripted
  • Conversational analytics copilot replacing manual BI dependency — natural-language querying of live logistics dashboards
  • Self-improving feedback loop — every outcome logged back, system learns from each resolution
02Build
Product Outline
  • Event trigger from structured logistics DB → orchestrator routing → parallel agent dispatch
  • Humanoid voice agent, analytics copilot, execution agent running concurrently against the same event
  • Auto-logged CRM + feedback layer closing the loop on outcomes and learnings
  • Modular AI backbone designed for full agentic execution as the next phase
03Outcome
Delivery & Impact
  • 1,000+ AI calls processed daily — directly increasing client revenue via automated feature upsell
  • Manual control-tower decisions replaced by automated decision intelligence
  • Ongoing active engagement — current project, expanding to full agentic execution
StackCrewAILangChainElevenLabs Voicen8nSQL + LLMEvent StreamingCRM Integration
1K+
AI calls per day
3
Agent types running in parallel
Teams
Control tower staff replaced
07
Brand & CampaignAI Rank PredictorEdTech
Marketing Campaign — Resonance Eduventures
Resonance (go, pal!) · May – Aug 2022
ReadClose
Idea
How do you build a brand campaign for one of India's largest competitive exam institutes — with a team of 8 IIT students?
Company
Resonance Eduventures is a major competitive exam coaching brand based in Kota. They needed a full brand and marketing campaign for a course launch — including offline presence, digital assets, and student engagement during JEE exam season.
01Approach
Suggested Solve
  • Full brand design, course-launch campaign, canopies, banners, jingles, promotional videos, and website hygiene
  • Built a rank predictor significantly better than existing tools in the competitive exam sector
  • Planned offline events for 1,000+ audiences; water and rest stalls outside JEE exam centres in Kota
02Build
Product Outline
  • Primary research across all stakeholders — students, parents, employees — to surface operational insights
  • 8-person IIT student team managing end-to-end delivery, paid above market internship standards
  • Operational efficiency and infrastructure cost-saving recommendations delivered alongside campaign
03Outcome
Delivery & Impact
  • Full campaign delivered across digital and offline channels
  • Rank predictor built and deployed — outperformed competitors in accuracy
  • Operational insights delivered to leadership; early proof of gopal's team-led consulting capability
StackAI Rank Prediction ModelPrimary ResearchBrand DesignVideo ProductionEvent Operations
1K+
Audience at events
#1
Rank predictor accuracy vs competitors
8
IIT students led — above-market pay
08
System DesignGovTechEmployment AI
Employment System for JanSuraaj Youth Clubs
JanSuraaj / Prashant Kishor (go, pal!) · Apr – May 2023
ReadClose
Idea
Design a scalable employment and skill-tracking system for blue-collar workers across Prashant Kishor's youth clubs.
Company
Prashant Kishor — the political strategist who brought data science to Indian politics — needed a system to help members of his JanSuraaj youth clubs find employment. The challenge: grading diverse skill sets, matching them to industrial opportunities, and building training pathways.
01Approach
Suggested Solve
  • Multi-level grading system for blue-collar workers and industrial operators
  • Educational modules, skill tracking, and opportunity mapping flows built into one system
  • Planned collaborations with industrial HR teams to align filtration and training to real hiring needs
02Build
Product Outline
  • Skill assessment and level-grading engine tailored to blue-collar and technical roles
  • Opportunity mapping: matching candidate profiles to live industrial requirements
  • Training pathway design: modules to upskill candidates to meet specific employer standards
03Outcome
Delivery & Impact
  • System designed and handed off to PK's JanSuraaj organisation
  • Framework applicable to lakhs of youth club members across Bihar
  • Experience working directly with Prashant Kishor on a politically and socially high-impact brief
StackSkill Assessment LogicOpportunity MatchingSystem DesignEducational Module Framework
L+
Youth club members in scope
PK
Direct collaboration with Prashant Kishor
Bihar
State-scale employment system designed
09
DeliveryTeam LeadershipMartech
Team Scale-Up — From Operator to Leading the Martech Practice
PivotRoots · CSA (Havas) · Apr 2025 – Feb 2026
ReadClose
Idea
Take a martech delivery team and make it bigger, faster and more profitable at once — without dropping client retention or blowing the revenue-to-expense ratio.
Company
CSA, part of Havas Media Network, runs marketing-technology delivery for a roster of enterprise brands. Parva moved from running owned-channel execution to leading the practice — with a mandate to scale the team, the revenue and the client book simultaneously while holding quality.
01Approach
Suggested Solve
  • Productise repeatable delivery — CDP/CEP deployment and owned-channel journeys — so the team scales without bespoke rework per account
  • Senior-led pods with named owners on a weekly loop, instead of a deep staffing pyramid
  • Protect retention by instrumenting outcomes per client, not just shipping deliverables
02Build
Product Outline
  • Reusable deployment playbooks across verticals and business cases
  • Hiring and onboarding to triple headcount while keeping revenue-to-expense in line
  • Account-expansion motion that grew the roster fourfold
03Outcome
Delivery & Impact
  • Team grown 3× · team revenue 2.5×
  • Client roster 4× · retention to 75%
  • Revenue-to-expense maintained through the scale-up
StackCDP / CEP DeploymentOwned-channel CRMDelivery PlaybooksTeam Ops
Team size grown
2.5×
Team revenue grown
75%
Client retention · roster 4×
10
Owned ChannelsCLMMulti-Vertical
12-Brand Owned-Channel Portfolio — One Operator, Many Funnels
PivotConsult (Havas) · Jan 2024 – Mar 2025
ReadClose
Idea
Run owned-channel strategy and execution for a dozen brands at once — QSR, D2C, BFSI and pharma — without each one becoming a special case.
Company
As Marketing Automation Specialist at PivotConsult, Parva drove sales and retention across 12+ brands simultaneously — KFC, Bisleri, Tata 1mg, FnP, HDFC PayzApp, Wildcraft, Woodland, Nysaa, IOB, House of Masaba, Grown Brilliance and Sleepwell.
01Approach
Suggested Solve
  • Shared CLM patterns across brands, calibrated per vertical — not twelve bespoke programs
  • Convert one-off campaigns into lifecycle journeys that keep working after launch
  • Segment on behaviour and recency so each brand's spend lands on the right cohort
02Build
Product Outline
  • Owned-channel journeys — email, SMS, push and WhatsApp — on each brand's CEP
  • RFM / behavioural cohorting with offer-and-comms laddering per segment
  • Campaign-to-journey conversion so launches compound instead of resetting
03Outcome
Delivery & Impact
  • 12+ brands run concurrently across four verticals
  • Recognised at the Smarties, Abby's and ET Martech+
  • The roster that anchored the later practice-lead mandate
StackBraze / CEPLifecycle JourneysRFM SegmentationCampaign-to-Journey CLMWhatsApp / Push / SMS
12+
Brands run concurrently
4
Verticals · QSR · D2C · BFSI · pharma
3
Industry awards · Smarties · Abby's · ET
11
AttributionMarketing ScienceOffline + Online
Spend-to-Outcome — Where ₹10Cr of Budget Becomes ₹3Cr of Result
PivotConsult / CSA (Havas) · 2024 – 2025
ReadClose
Idea
Most dashboards report spend, not outcome. Build the attribution layer that shows exactly where the budget leaks — across offline and online — and what actually drove the number.
Company
Brands pour crores into a mix of TV, OOH, print, Google, Meta and LinkedIn, then credit whatever the last-click dashboard happens to see. Parva built a bespoke attribution engine that modelled organic and social headwinds across offline and online spend, isolating incremental contribution so the gap between budget in and outcome out could finally be read.
Architecture — Spend-to-Outcome · Headwinds Removed, Incremental Isolated
Offline spend · TV/OOH/Print Online · Google/Meta/LinkedIn Organic + social baseline GA4 · Sales · CRM ATTRIBUTION MODEL Strip organic + social headwinds Isolate paid incremental Offline + online reconciled INCREMENTAL LIFT Wasted Working TRUE SPEND→OUTCOME What actually drove the number REALLOCATION MAP Move budget to what works ₹10Cr → ₹3Cr the leak, made visible ◆ SPEND-TO-OUTCOME — organic/social headwinds removed · paid incremental isolated · offline + online reconciled
◆ Spend-to-Outcome Attribution
01Approach
Suggested Solve
  • Model the organic and social baseline so paid effect isn't credited with background demand
  • Isolate paid incremental from headwinds and tailwinds the dashboard can't see
  • Reconcile offline and online into a single spend-to-outcome view, not two siloed reports
02Build
Product Outline
  • Ingest media spend, GA4 and sales/CRM into one modelled view
  • Baseline + headwind model, then incrementality scoring per channel
  • Leak surfacing with a reallocation map — where to move the next rupee
03Outcome
Delivery & Impact
  • Read the number nobody else could read — true incremental, not last-click
  • Surfaced where ₹10Cr of budget produces ₹3Cr of outcome
  • Reallocation guidance the team could act on the same week
StackAttribution ModelingGA4 Data APIOffline + Online SpendPythonIncrementality
₹10→3Cr
Budget-to-outcome leak surfaced
On+Off
Offline + online reconciled
Reallocation the team could action
12
CDPReal-Time DecisioningRetail
AI Inventory Tool — Real-Time Decisioning in Front of the Team
CSA (Havas) · 2024 – 2025
ReadClose
Idea
Put real-time decisioning in front of the marketing team — what to do next, now — instead of a report that lands a week late.
Company
Alongside a CDP build, Parva shipped an AI inventory / decisioning tool that surfaced the next best action in real time rather than after the fact — recognised at the Smarties. (Flag: internal architecture and hard metrics for this tool aren't documented here yet — this is the grounded outline; send specifics to deepen it.)
01Approach
Suggested Solve
  • Resolve identity in a CDP so decisions run on a single, current view of the customer
  • Move decisioning from weekly reporting to in-the-moment recommendation
  • Surface the call where the team already works, not in a separate BI tool
02Build
Product Outline
  • CDP + identity resolution as the data spine
  • Real-time decisioning layer feeding next-best-action to the team
  • Inventory / availability signals folded into the recommendation
03Outcome
Delivery & Impact
  • Real-time decisioning live in front of the team
  • Recognised at the Smarties
  • Part of the CDP-plus-AI stack behind the practice's results
StackCDPIdentity ResolutionReal-Time Decisioning
RT
Real-time, not week-late
Smarties recognition
CDP
Identity-resolved decisioning
13
GTMAI ProductVenture
dotler.ai — AI-Native GTM, Built and Exited in a Quarter
dotler.ai · Co-Founder · Feb – Apr 2026 · Bengaluru
ReadClose
Idea
Solve GTM, workflow and martech problems with AI — and find out fast whether the market is really there.
Company
Parva co-founded dotler.ai in Bengaluru — agent architecture and product build over a focused three-month sprint, then a clean exit once the research showed where the market wasn't. (Flag: product internals are summarised from the founder brief — send specifics to expand the build section.)
01Approach
Suggested Solve
  • Agent architecture aimed at the GTM and workflow grunt-work martech teams hate
  • Build fast, ship to real operators, and let the market — not the deck — decide
  • Treat the venture as a falsifiable bet with a short clock
02Build
Product Outline
  • AI-native GTM / workflow product with an agentic core
  • Three-month build-and-learn sprint with live operator feedback
  • Outreach and discovery to pressure-test demand
03Outcome
Delivery & Impact
  • Product built and shipped inside a quarter
  • Clean exit — found where the market actually wasn't, fast
  • The pattern that fed straight back into go, pal!
StackAgent ArchitectureWorkflow AutomationGTM
3mo
Build-and-learn sprint
Exit
Clean, fast
Learnings fed into go, pal!
14
AI InfraMCPLogistics
Fleet-Tech AI Build — MCP over Live Telematics
go, pal! · 2026 – Present
ReadClose
Idea
Give a logistics client's team live, queryable intelligence over their fleet — and a lead pipeline that screens itself — on infrastructure that stays auditable and dependency-light.
Company
gopal's first delivery: an AI + analytics build for a logistics / fleet-telematics client. Reporting, live vehicle-data access for agents, and self-screening lead intake — built so the data layer is owned before the agent layer, and nothing depends on a black box.
Architecture — Owned Data Layer · MCP Tool Surface · Self-Screening Intake
A · REPORTING B · LIVE OPS C · INTAKE Google Ads LinkedIn Ads GA4 GITHUB ACTIONS ETL Scheduled pulls Cron · idempotent SUPABASE POSTGRES Owned data spine Single source of truth REACT + SHADCN Reporting dashboard For the client's team Fleet Telematics APIs Live vehicle data MCP SERVER Telematics as a tool surface Typed, auditable calls AGENTS Query live vehicle data Grounded in real state Inbound leads VOICE AGENT Bilingual · Hindi / English CLASSIFIED JSON Structured lead records SCOPED · NEXT viAct + Roboflow Vision layer On-site compliance ◆ OWN THE DATA LAYER BEFORE THE AGENT LAYER — auditable, dependency-light, production-first
◆ MCP Tool Surface
◆ Owned Data Layer
01Approach
Suggested Solve
  • Own the data layer — ETL into Postgres — before building anything agentic on top
  • Wrap the telematics APIs as an MCP tool surface so agents query live vehicle data through typed, auditable calls
  • Automate lead screening so intake classifies itself instead of waiting on a human
02Build
Product Outline
  • React / shadcn reporting dashboard on Supabase Postgres, fed by a GitHub Actions ETL pulling Google + LinkedIn Ads and GA4
  • MCP server over the fleet telematics APIs — agents query live vehicle state
  • Bilingual Hindi/English voice screening agent classifying leads to JSON; viAct + Roboflow vision layer scoped for on-site compliance
03Outcome
Delivery & Impact
  • Dashboard + ETL + MCP layer live; agents query live fleet data
  • Bilingual voice intake turning calls into structured JSON
  • Zero-dependency, auditable build — engagement ongoing (no revenue figure to report yet)
StackReactshadcn/uiSupabase / PostgresGitHub Actions ETLMCP ServerGA4 · Ads APIsVoice AIviAct · Roboflow
Live
Dashboard · ETL · MCP in production
Hi/En
Bilingual voice → structured JSON
0-dep
Auditable, production-first
09
Emotional AINeuroplasticityChain-of-Listening
Unstress.ai — India’s Missing Emotional Infrastructure
Self-Founded · Feb 2025 – Present
ReadClose
Idea
95% of India's mental health sufferers never reach care. Build an AI companion that meets them where they are — WhatsApp-native, shame-free — and rewires stress patterns subconsciously.
Company
India's mental health spend is 10× lower than western averages. The gap isn't just affordability — it's cultural design. Therapy feels elite. Advice feels intrusive. Emotional learning is never taught. Puneet spent three years in therapy research, shadow work, and reiki before founding Unstress. Built the Chain-of-Listening (CoL) engine for stress-pattern rewiring with a WhatsApp-first, emotionally aware UX.
Architecture — Chain-of-Listening · Stressor-to-Therapy Progression Engine
CONVERSATION Normal chat User unaware of depth DEEP ANALYSIS Sentiment · Tone · Triggers CBT · Vedanta frames Psychoanalytic signals Pattern · Stage detection STRESSOR → THERAPY STAGE MAP Problem identified Therapy path assigned Stage progress tracked All hidden from user READINESS-GATED INTERVENTION Fires only when user is truly open RAG · ChromaDB Therapy content retrieval Session memory layered DYNAMIC PROMPT Personalised · Reframed Never intrusive Response Earned · Timed Stage progress update · Neuroplasticity loop ◆ CHAIN-OF-LISTENING — Stressor-to-therapy mapping hidden under normal conversation
◆ Chain-of-Listening (CoL)
◆ Readiness-Gated Intervention
◆ Stressor-to-Stage Mapping
01Approach
Suggested Solve
  • Chain-of-Listening engine: track emotional tone and patterns, calculate readiness scores, intervene only when the user is receptive — never pushy
  • Story-driven insights delivered at receptive moments using CBT and Vedanta frameworks — not clinical advice
  • WhatsApp-native, shame-free, culturally calibrated for India — designed to feel like someone who gets you
02Build
Product Outline
  • Chat analysis → sentiment / tone / trigger detection → Solution Tracker → RAG content retrieval → dynamic personalised prompt
  • Readiness scoring: reflection mode only activates when the user is open — never intrusive
  • Emotional filters simulating softness, hesitation, non-repetitive tone for authentic human-like connection
  • Tiered pricing: Freemium → ₹49 → ₹1,800 / yr → ₹4,800 / yr deep support
03Outcome
Delivery & Impact
  • 100+ waitlist users pre-MVP launch
  • 8 / 10 target users already use GPT for personal issues — proven demand for AI emotional support
  • Early VC interest; corporate beta in preparation
  • Category creation: culturally designed emotional AI for India
StackLangChainChromaDBRAG + Session MemoryOpenAI APIWhatsApp APICBT + Vedanta Logicn8n
95%
Indian sufferers who never reach care — the TAM
100+
Waitlist users pre-launch
0
Visible therapy — all changes happen beneath the surface

Most enterprise teams think in old frameworks and don't know what agents can actually do. The opportunities below are where AI replaces entire functions — not assists them. These are the verticals with the highest leverage, highest billing, and clearest ROI for clients.

M / 01
Enterprise Marketing Automation
Learn a client's marketing SOPs, extract first principles, embed that reasoning into agents that run the entire function. Audience segmentation, campaign creation, geo-targeted execution, and channel launch in a single session. Reduces team size to 30% or multiplies output 5×.
Campaign AgentsBrand IntelligenceWhatsApp / OmnichannelGeoIQ IntegrationAgentic Dashboard
R / 02
Deep Research & Compliance Agents
Any function that requires researching a person, entity, or topic across multiple sources — compliance, due diligence, media vetting, legal audits — can be automated with confidence-scored, audit-ready output. PEP compliance was just one example. The pattern applies everywhere.
Multi-Source ResearchConfidence ScoringAudit ReportsLegal VettingMedia Audits
V / 03
Humanoid Voice AI Systems
Sales SDRs, appointment setters, customer care, operations communication — all replaceable with context-aware, emotionally-calibrated voice agents. Not IVR. Not scripted bots. Adaptive, humanoid, capable of full conversations at any scale.
Outbound SalesCustomer CareAppointment SettingOps CommunicationElevenLabs
I / 04
Institutional Memory & Meeting Intelligence
Board meetings, leadership reviews, investor calls — all captured, indexed, and made queryable. The agent participates in live meetings, flags contradictions with prior decisions, and ensures nothing is lost between sessions. Deployed for Dubai Government. Interest from Adani Group.
Board Meeting AgentNER ExtractionContradiction DetectionVector + Graph MemorySecure · Encrypted
S / 05
SOP-to-Agent Conversion
Any team running on documented SOPs — engineering, operations, finance, HR — can have those SOPs converted into agent systems that execute the work. Claude Code and similar tools are generic; this is custom agent infrastructure built around a specific organisation's processes and reasoning.
SOP IngestionAgentic FrameworkDept-Specific AgentsHLSD DesignAEO
D / 06
Domain-Specific Algorithm Design
Any domain with structured rules, deep combinatorics, or proprietary logic — astrology, logistics routing, financial modelling, medical protocols — can have a custom symbolic-neural hybrid built on top of it. RAG alone is not enough. The real moat is in the rule engine beneath.
Symbolic AIRule Engine + LLMSQL + Graph + RAGDomain AlgorithmsToken Optimisation
Open to

All data retrieval modes: RAG · Graph · SQL · MCP. Any complexity of agent system. Product design for any of these verticals. Humanoid communication where voice replaces teams. High-billing enterprise engagements where AI saves significant human hours daily. Happy to come with ideas — most client teams don't know what's possible yet.

F · 012023 →
P
IIT DelhiDelivery
Delivery
Parva
Runs the engagements. Builds the schedule, the brief, the room.
Read bio
F · 022023 →
A
BITS · NITIE MumbaiMartech
Martech
Abhimanyu
Turns board-room ambition into deployed systems — CDPs, Gen-AI, predictive analytics.
Read bio
F · 032023 →
P
IIT BombayResearch
Research
Puneet
Lives on the model frontier. And in the unglamorous work of evaluating it.
Read bio
F · 042023 →
V
IIT DelhiBuilder
Builder
Vinayak
Takes the plan to prototype, then productionises the impact we promised.
Read bio
Now booking · Q3 2026

Build something deliberately.

Two engagement slots open. Best fits: enterprise marketing automation, voice ops, compliance research, board/meeting intelligence, custom symbolic-neural systems. Bring a problem, not a brief.

Schedule
Pick a slot — 30 minutes, no prep required
Open in new tab ↗