Aditya Hegde Bengaluru, India

Product manager who ships AI systems — from a PhD in guidance and control to production GenAI.

10+ years across automotive, aerospace, and AI. Currently building regulatory-AI tools; previously took a GenAI SaaS product from requirements to production.

hegdem.aditya@gmail.com LinkedIn GitHub ClearFlight — live app →
01Case study

Intellia

Confidential engagement

A GenAI product for turning scattered market intelligence into answers analysts could trust — taken from requirements to production.

The problem

Users needed one trustworthy answer that often lived across two different kinds of data — structured records and unstructured documents — and "trustworthy" had no agreed definition yet. Early feedback was "this feels off," with no way to say why.

What I did
  • Led delivery of an agentic retrieval system that routed each query to the right mix of data sources and search strategies, rather than forcing every question through one path.
  • Built a 189-question evaluation set — routing accuracy, fact/entity recall, LLM-as-judge — that turned "feels off" into a specific, fixable number, and turned stakeholder disagreements into resolvable technical ones.
  • Ran the cross-functional loop translating business requirements into architecture decisions.
Shipped to production. Cut manual curation effort by 30% via automated tagging and summarization. The evaluation framework became the team's standing quality bar, not a one-off audit.
query in router structured records both unstructured documents merge · rank answer out eval 189 Qs
The retrieval pipeline. The evaluation framework samples both the router and the merged answer, feeding flagged cases back into the router — the loop that replaced subjective QA.
02Case study

ClearFlight

Live app

An LLM tool that turns DGCA drone-certification clauses into a filterable traceability matrix. Open the live app →

The problem

Certification clauses are dense regulatory text. Engineers need to trace which requirement maps to which design element — get it wrong once and the cost is a failed audit, not a bug report.

What I did
  • Scoped the tool with a discovery interview with a UAV certification engineer, rather than guessing at the schema.
  • Built and deployed the LLM pipeline that extracts clauses into structured, filterable rows.
  • Reviewed all 17 rows against source text myself; 6 were flagged as wrong or ambiguous. Converted each failure into a schema constraint, a validator, or a prompt rule.
Live, working app — anyone can open it and query the matrix directly.

View the repo on GitHub →

clause text LLM extraction structured row human review vs. source pass matrix fail — 6 of 17 constraint · validator or prompt rule
The loop that hardened the pipeline: a failed row doesn't just get corrected once — it becomes a rule that extraction obeys from then on.
03Career timeline

Mechanical systems to AI systems

Nine years of controls and GNC engineering, a PhD in multi-agent guidance, then product.

2006–10 BE Mech. Eng. Delhi University Formula Student, Silverstone 2010–12 MS Mech. Eng. Univ. of Michigan Roll-stability research 2 patents 2012–15 Controls Engineer Cummins DEF dosing strategy 2015–17 Sr. Systems Integ. Cummins −20% downtime 2017–22 PhD Aerospace Eng. IISc Bengaluru Multi-agent UAV guidance 2022–23 Lead GNC Engineer General Aeronautics DGCA certification, 11 eng. 2024 Assoc., Decision Sci. Datazymes HR automation tool 2024–26 Sr. Mgr, Product Datazymes ↳ Intellia 2026– ClearFlight Independent Live app ↑
04How I work

What carries across every project

01

Measure quality before arguing about it

Evaluation frameworks, not opinions, settle whether something is good enough — at Datazymes and at ClearFlight alike.

02

Scope with the person who'll feel it break

A certification engineer for ClearFlight, cross-functional stakeholders for Intellia — the schema comes from a discovery conversation, not a guess.

03

Turn one failure into a system change

Six flagged rows became validators; a "feels off" became a 189-question benchmark. The fix is never just the one instance.

05Work experience

Experience

Senior Manager, Product Management Apr 2024 – Jul 2026
Datazymes Analytics Pvt. Ltd., Bengaluru, India
  • Product delivery: led the requirements-to-production launch of a GenAI market-intelligence SaaS for pharma — an agentic RAG system with hybrid retrieval (BM25 + vector) and multi-layer query routing across structured and unstructured data.
  • Quality and evaluation: established a 189-question evaluation framework (routing accuracy, fact/entity recall, LLM-as-judge), replacing qualitative QA with a measurable baseline.
  • Outcome and execution: reduced manual curation effort by 30% via automated AI tagging and summarization; ran cross-functional sprints translating business requirements into technical roadmaps.
  • Commercial and client work: ran demos and requirement-gathering with 3–4 prospective clients; contributed to the pricing model for a competitive RFP, which we won.
Associate, Decision Sciences Jan 2024 – Apr 2024
Datazymes Analytics Pvt. Ltd., Bengaluru, India
  • Internal optimization: architected an HR automation tool reducing process latency from 2 days to 5 minutes; developed Streamlit dashboards for real-time resource tracking.
Lead GNC Engineer Dec 2022 – Nov 2023
General Aeronautics Pvt. Ltd., Bengaluru, India
  • Systems leadership: directed 11 engineers to architect flight control systems for autonomous drones; secured 100% DGCA Type Certification for two flagship product lines.
  • R&D and ops: led FMCW radar/LiDAR sensor selection and obstacle-avoidance algorithm development; migrated the team to standardized Bitbucket CI/CD with a monthly release cycle.
Senior Systems Integration Engineer Mar 2015 – Jul 2017
Cummins Inc., Indiana, USA
  • Customer ROI: defined and shipped a fault-handling architecture addressing vehicle downtime and warranty costs, reducing downtime by 20% and catching 100% of production-line installation failures.
  • SME and enablement: served as OBD Subject Matter Expert; translated regulatory requirements into actionable standards, coaching engineers on compliance and system-integration processes.
Controls Engineer Jul 2012 – Mar 2015
Cummins Inc., Indiana, USA
  • IP and quality: defined and patented a unified DEF injection strategy across multi-dosing architectures (2 US patents); applied FMEAs and P-diagrams for the QSK95 engine series.
06Academic

Publications, review, and teaching

Publications — Journals

  1. A. Hegde and D. Ghose, "Collaborative Transportation with Multi-rotor UAVs Using Distributed and Non-smooth Control." Submitted to a journal.
  2. A. Hegde and A. Jain, "Synchronization and Balancing around Simple Closed Polar Curves with Bounded Trajectories and Control Saturation." Automatica, vol. 149, p. 110810, 2023.
  3. A. Hegde, J. J. Aloor, and D. Ghose, "Bounded Distance-control for Multi-UAV Formation Safety and Preservation in Target-tracking Applications." Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, 2022.
  4. A. Hegde and D. Ghose, "Multi-quadrotor Distributed Load Transportation for Autonomous Agriculture Spraying Operations." Journal of Guidance, Control, and Dynamics, vol. 45, no. 5, pp. 944–951, 2022.
  5. A. Hegde and D. Ghose, "Multi-UAV Collaborative Transportation of Payloads With Obstacle Avoidance." IEEE Control Systems Letters, vol. 6, pp. 926–931, 2022.

Publications — Conferences

  1. A. Hegde and D. Ghose, "Collaborative Guidance of UAV-Transported Semi-Flexible Payloads in Environments with Obstacles." 60th IEEE Conference on Decision and Control (CDC), 2021, pp. 490–495.
  2. A. Hegde and D. Ghose, "Multi-UAV Distributed Control for Load Transportation in Precision Agriculture." AIAA SciTech Forum, 2020.

Peer review

  • Reviewer for IEEE Transactions and Letters journals: IEEE L-CSS, IEEE TAES, IEEE ITS
  • Reviewer for IEEE conferences: IEEE CCTA, IEEE CDC, IEEE ACC
  • Reviewer for IFAC journal and symposium: Automatica, and the IFAC Symposium on Automatic Control in Aerospace (2022)

Research

Research Assistant, University of Michigan, Ann Arbor (2011)

  • Optimized roll stability controllers for a vehicle in CarSim and MATLAB/Simulink for a double-lane change maneuver.
  • Used Pareto analysis to balance competing roll stability and track-following objectives.

Teaching and training

Teaching Assistant, Mathematical Methods for Aerospace Engineers (2022), IISc Bengaluru

Trainer, Guidance of Autonomous Vehicles (2019), IISc Bengaluru

  • Led programming sessions for Bharat Electronics Ltd. employees on mathematical modeling of dynamic systems in MATLAB and on implementing guidance and navigation algorithms for autonomous vehicles.

Student leadership

  • Chair, IEEE Student Branch (2021–22) — Indian Institute of Science. Organized technical, professional, and networking programs; served as liaison to the IEEE-Bangalore Section.
  • President, IEEE-HKN Mu-Xi Chapter (2020–22) — Indian Institute of Science. Led professional and networking activities and member inductions for the Mu-Xi chapter of the IEEE Eta Kappa Nu (HKN) academic honor society.